Method for generating sample palm print images, and apparatus, devices, and computer programs

Abstract

A computer-implemented method for generating a sample palmprint image is provided, the method comprising: generating at least one palmprint main line (210) based on a predetermined palmprint main line distribution rule; determining a palmprint distribution region in which the at least one palmprint main line is located, and generating at least one palmprint thin line within the palmprint distribution region based on a predetermined palmprint thin line distribution rule (220); generating a candidate palmprint image based on the at least one palmprint main line and the at least one palmprint thin line within the palmprint distribution region, wherein the candidate palmprint image is labeled with a candidate image label, the candidate image label being used to indicate a palmprint distribution of the candidate palmprint image, the palmprint distribution including at least one of a palmprint main line distribution and a palmprint thin line distribution (230); and inputting the candidate palmprint image into a pre-trained palmprint generation model to obtain a sample palmprint image (240).

Description

[Technical Field]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This application claims priority to a Chinese patent application filed on April 19, 2013, bearing application number 2023104648718 and entitled "Method for generating palm print samples and apparatus, devices, medium, and program products thereof," the entire contents of which are incorporated herein by reference.

[0002] The embodiments of the present application relate to the field of machine learning, and in particular to a method for generating a sample palm print image, an apparatus, a device, a medium, and a program product thereof. [Background technology]

[0003] With the rapid development of information technology, palmprint recognition technology is increasingly being used in various identity authentication scenarios due to its reliability and convenience. Palmprint recognition is based on the characteristics of the palm, such as the main lines, patterns, and wrinkles, and is a non-invasive identification method compared to face recognition, which is more easily accepted by users.

[0004] In related art, when identifying palmprints, machine learning methods are usually used to learn the inherent rules of palmprint information in collected palmprint images, and the model is trained to learn discriminatory latent features, and the palmprint is identified using the trained model to realize the identity recognition process.

[0005] However, due to the high privacy nature of palmprint images, there is a lack of large public datasets for training models in the field of palmprint recognition, and the unreliability of randomly generated palmprint samples makes it difficult for models to learn better identity information, resulting in a low effectiveness of the model's identity recognition. Summary of the Invention

[0006] The present application provides a computerized method for generating a sample palm print image, the method comprising: generating at least one palm print main line based on a predetermined palm print main line distribution rule; determining a palm print distribution area in which the at least one palm print main line is located, and generating at least one palm print thin line within the palm print distribution area based on a predetermined palm print thin line distribution rule; generating a candidate palm print image based on the at least one palm print main line and the at least one palm print thin line in the palm print distribution region, wherein the candidate palm print image is labeled with a candidate image label, the candidate image label is used to indicate a palm print distribution of the candidate palm print image, and the palm print distribution includes at least one of a palm print main line distribution and a palm print thin line distribution; and inputting the candidate palm print image into a pre-trained palm print generation model to obtain a sample palm print image, the palm print generation model being a model pre-trained using palm print image labels of pre-collected palm print images and the candidate image labels, the palm print image labels being used to indicate the palm print distribution of the pre-collected palm print images, and the sample palm print image being used to train a palm print recognition model.

[0007] In another aspect, there is provided an apparatus for generating a sample palm print image, the apparatus comprising: a main line generating module configured to generate at least one palm print main line based on a predetermined palm print main line distribution rule; a thin line generating module configured to determine a palm print distribution area in which the at least one palm print main line is located, and to generate at least one palm print thin line within the palm print distribution area based on a predetermined palm print thin line distribution rule; an image acquisition module configured to generate a candidate palm print image based on the at least one palm print main line and the at least one palm print thin line in the palm print distribution region, wherein the candidate palm print image is labeled with a candidate image label, the candidate image label is used to indicate a palm print distribution of the candidate palm print image, and the palm print distribution includes at least one of a palm print main line distribution and a palm print thin line distribution; an image generation module configured to input the candidate palm print image into a pre-trained palm print generation model to obtain a sample palm print image, the palm print generation model being a model pre-trained using palm print image labels of pre-collected palm print images and the candidate image labels, the palm print image labels being used to indicate a palm print distribution of the pre-collected palm print images, and the sample palm print image being used to train a palm print recognition model.

[0008] In another aspect, there is provided a computing device comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the processor loads and executes the at least one instruction, the at least one program, code set or instruction set to perform the method for generating a sample palmprint image as set forth in any one of the above embodiments of the present application.

[0009] In another aspect, there is provided a computer-readable storage medium storing at least one instruction, at least one program, code set or instruction set which, when loaded and executed by a processor, causes the processor to perform the method for generating a sample palmprint image as set forth in any one of the above embodiments of the present application.

[0010] In another aspect, there is provided a computer program product or a computer program comprising computer instructions stored in a computer-readable storage medium, wherein a processor of a computing device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computing device to perform the method for generating a sample palmprint image according to any one of the above embodiments of the present application.

[0011] The details of one or more embodiments of the application are set forth in the drawings and description which follow. Other features, objects, and advantages of the application will become apparent from the description, drawings, and claims. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a schematic diagram of an implementation environment according to one exemplary embodiment of the present application; [Figure 2] 1 is a flowchart of a method for generating a sample palm print image according to one exemplary embodiment of the present application. [Figure 3] 1 is a schematic diagram of a palm print according to one exemplary embodiment of the present application; [Figure 4] FIG. 2 is a schematic diagram of a candidate palm print image according to one exemplary embodiment of the present application; [Figure 5] 1 is a flowchart of obtaining a palmprint generation model by training according to one exemplary embodiment of the present application; [Figure 6] 1 is a flowchart of a method for generating a sample palm print image according to another exemplary embodiment of the present application. [Figure 7] FIG. 2 is a schematic diagram of a Bezier curve according to one exemplary embodiment of the present application; [Figure 8] 10 is a flowchart of a method for generating a sample palm print image according to yet another exemplary embodiment of the present application. [Figure 9] 10 is a flowchart of a method for generating a sample palm print image according to yet another exemplary embodiment of the present application. [Figure 10] 1 is a flowchart of applying a target palm print recognition model according to one exemplary embodiment of the present application; [Figure 11] 1 is an overall flowchart of a method for generating a sample palm print image according to one exemplary embodiment of the present application. [Figure 12] FIG. 2 is a schematic diagram of a third region according to one exemplary embodiment of the present application. [Figure 13] FIG. 10 is a schematic diagram of determining a third region according to one exemplary embodiment of the present application; [Figure 14] FIG. 2 is a schematic diagram of a candidate palm print image according to one exemplary embodiment of the present application; [Figure 15] FIG. 1 is a principle diagram of a model of the training stage of a generative adversarial network according to one exemplary embodiment of the present application; [Figure 16]FIG. 2 is a schematic diagram of a network structure of a decoder according to one exemplary embodiment of the present application; [Figure 17] FIG. 1 is a schematic diagram of obtaining a target palm print recognition model by training according to one exemplary embodiment of the present application; [Figure 18] FIG. 1 is a block diagram of the structure of a sample palm print image generation device according to one exemplary embodiment of the present application. [Figure 19] FIG. 10 is a block diagram of the structure of a sample palm print image generation device according to another exemplary embodiment of the present application. [Figure 20] FIG. 2 is a block diagram of a server architecture according to one exemplary embodiment of the present application; DETAILED DESCRIPTION OF THE INVENTION

[0013] In order to make the technical solutions in the embodiments of the present application clearer, the above briefly introduces drawings necessary for describing the embodiments. The drawings in the above description are only some embodiments of the present application, and it is obvious to those skilled in the art that other drawings can also be obtained according to these drawings without any creative work.

[0014] The following clearly and completely describes the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application, but it is clear that the described embodiments are only a part of the embodiments of the present application, and are not all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present application.

[0015] In the related art, when identifying palmprints, machine learning methods are usually used to learn the inherent rules of palmprint information in collected palmprint images, train a model to learn discriminatory latent features, and then use the trained model to identify palmprints, thereby realizing the process of identity recognition. However, because palmprint images have a high level of privacy, there is a lack of large public datasets for models to learn in the field of palmprint recognition, and the reliability of randomly generated palmprint samples is low, which makes it difficult for models to learn better identity information, resulting in a low effectiveness of the model in identity recognition.

[0016] In the embodiments of the present application, a method for generating sample palmprint images is provided, which obtains a large number of diverse candidate palmprint images, and then uses pre-collected palmprint images to perform constraint adjustment on the generated candidate palmprint images, thereby improving the reliability of the generated sample palmprint images and helping to improve the robustness of the palmprint recognition model. The method for generating sample palmprint images obtained by training in the present application includes various scenarios when applied, such as palmprint recognition model training scenarios, palmprint encryption scenarios, and palmprint recognition scenarios. It should be noted that the above application scenarios are merely exemplary, and the method for generating sample palmprint images provided in the present application can also be applied to other scenarios, and the embodiments of the present application are not limited thereto.

[0017] It should be noted that the present application may display a presentation interface, a pop-up window, or output audio presentation information before and during the process of collecting user-related data, and the presentation interface, pop-up window, or audio presentation information is used to notify the user that user-related data is currently being collected. The present application only starts to perform the relevant steps of acquiring user-related data after receiving a user confirmation action on the presentation interface or pop-up window; otherwise (i.e., if the user does not confirm the presentation interface or pop-up window), the relevant steps of acquiring user-related data are terminated, that is, the user-related data is not acquired. In other words, all user data collected by the present application is collected with the user's consent and permission, and the collection, use, and processing of related user data should comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0018] Next, an implementation environment according to the embodiment of the present application will be described. The sample palmprint image generation method according to the embodiment of the present application may be performed by a terminal alone, a server, or through data interaction between a terminal and a server, and the embodiment of the present application is not limited thereto. Optionally, the data transmission method performed through interaction between a terminal and a server will be described as an example.

[0019] For example, referring to FIG. 1, the implementation environment involves a terminal 110 and a server 120, and the terminal 110 and the server 120 are connected via a communication network .

[0020] In some embodiments, an application with a palmprint image acquisition function is installed on the terminal 110. Illustratively, the terminal 110 is used to transmit a palmprint image library to the server 120. The palmprint image library stores a plurality of collected palmprint images in advance. The server 120 can determine palmprint data information, such as a palmprint main line distribution rule and a palmprint thin line distribution rule, based on the palmprint image library.

[0021] Alternatively, an application having a palmprint image analysis function is installed in the terminal 110. For example, the terminal 110 collects a plurality of palmprint images, performs image analysis on the plurality of palmprint images, and then determines palmprint data information such as a palmprint main line distribution rule and a palmprint fine line distribution rule. Then, the terminal 110 transmits the palmprint main line distribution rule and the palmprint fine line distribution rule to the server 120.

[0022] In some embodiments, the server 120 generates at least one palmprint main line based on a predetermined palmprint main line distribution rule, determines a palmprint distribution region in which the at least one palmprint main line is located, generates at least one palmprint thin line within the palmprint distribution region based on a predetermined palmprint thin line distribution rule, and constructs a candidate palmprint image using the at least one palmprint main line and the at least one palmprint thin line, and assigns a candidate image label to the candidate palmprint image, the candidate image label representing the distribution of the palmprint main lines in the candidate palmprint image.

[0023] Optionally, the server 120 inputs the candidate palm print image into a pre-trained palm print generation model to obtain a sample palm print image, where the palm print generation model is a model pre-trained using palm print image labels of pre-collected palm print images and candidate image labels, and the palm print image labels are for indicating the palm print distribution of the palm print image.

[0024] In some embodiments, sample palmprint images obtained by processing candidate palmprint images with the palmprint generation model are used to train palmprint recognition model 121, which is then used to perform palmprint recognition. Because the palmprint generation model is obtained by training collected palmprint images, the sample palmprint images are numerous and highly accurate. By using the sample palmprint images to train palmprint recognition model 121, palmprints can be authenticated using the trained palmprint recognition model, improving the accuracy of palmprint recognition.

[0025] It should be noted that the above terminals include, but are not limited to, mobile terminals such as mobile phones, tablet computers, portable laptop computers, intelligent voice interaction devices, intelligent home appliances, and in-vehicle terminals, and may also be realized as desktop computers, etc. The above server may be an independent physical server, a server cluster or a distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), big data and artificial intelligence platforms, etc.

[0026] Here, cloud technology refers to hosting technology that integrates a series of resources, such as hardware, applications, and networks, within a wide area network or local area network to realize computing, storage, processing, and sharing of data. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, and application technology based on the business model application of cloud computing, which can form a resource pool that can be used as needed, flexibly, and conveniently.

[0027] In some embodiments, the above servers may be implemented as nodes in a blockchain system.

[0028] Combining the introduction of the above terms and application scenarios, the sample palmprint image generation method provided by the present application will be described. Taking the application of the method to a server as an example, as shown in FIG. 2, the method includes the following steps 210 to 240.

[0029] In step 210, at least one palm print main line is generated based on a predetermined palm print main line distribution rule.

[0030] Here, the palmprint main line distribution rule describes the distribution of the main lines in a palmprint and is used to indicate the distribution of the main lines in a palmprint. The palmprint main line distribution rule may include a position rule and a direction rule of the main lines in a palmprint, where the position rule describes the position of the main lines in a palmprint and the direction rule describes the tendency of the main lines in a palmprint to extend. The palmprint main line distribution rule may be obtained by comprehensively summarizing the palmprints of living organisms by artificial or artificial intelligence means. The palmprint main line distribution rule is also called the distribution mode or distribution method of palmprint main lines.

[0031] Palmprint fine line distribution rules describe the distribution of fine lines in palmprints and are used to indicate the distribution of fine lines in palmprints. Palmprint fine line distribution rules may include position rules and direction rules for fine lines in palmprints, where the position rules describe the positions of fine lines in palmprints and the direction rules describe the tendency of fine lines in palmprints to extend. Palmprint fine line distribution rules may be obtained by comprehensively summarizing palmprints of living organisms by artificial or artificial intelligence means. Palmprint fine line distribution rules are also called palmprint fine line distribution modes or palmprint fine line distribution methods.

[0032] Optionally, by analyzing the palm prints of many palms owned by the living being itself, it is determined that the palm prints mainly include main lines and fine lines, wherein the general characteristics of the palm prints include at least one of: (1) the distribution of the main lines is relatively regular, and the distribution of the fine lines is relatively random; and (2) the main lines are usually longer, thicker, and deeper in the palm print, and the fine lines are thinner, shorter, and shallower compared to the main lines.

[0033] Optionally, when the palm print is a palm print, the distribution of the palm print main lines usually shows a diagonal relationship, that is, the distribution rule of the palm print main lines is a diagonal rule. For example, in the case of a left palm, the palm print main lines generally start from the upper left corner and end at the lower right corner, and in the case of a right palm, the palm print main lines generally start from the upper right corner and end at the lower left corner. As shown in FIG. 3 , a schematic diagram of a palm print of a left palm is shown, where a palm area 310 includes main lines 320 and thin lines 330, with thicker lines representing the main lines 320 and thinner lines representing the thin lines 330, that is, the palm area 310 includes three main lines 320 and thirteen thin lines 330.

[0034] In some embodiments, at least one palmprint main line is generated based on a palmprint main line distribution rule in a predetermined palmprint main line quantity range.

[0035] For example, the range of the number of palm print main lines is determined by analyzing the palm prints of many palms owned by the living organism. For example, when observing based on palm print data owned by the living organism, the number of palm print main lines is generally 2 to 5.

[0036] In one alternative embodiment, the positioning point data is generated based on a predetermined palm print main line distribution rule.

[0037] Here, the positioning point data includes first data corresponding to the first main line positioning point and second data corresponding to the second main line positioning point.

[0038] The positioning point data is used to indicate the fixed status of the main line and includes first data and second data, and the start status and end status of the main line are determined based on the first data and the second data, and the distribution of the main line is approximately determined. Illustratively, the first data corresponds to a first main line positioning point, and the second data corresponds to a second main line positioning point.

[0039] Optionally, the starting point of the palm print main line is the first main line positioning point, and the ending point of the palm print main line is the second main line positioning point.

[0040] In one alternative embodiment, the adjustment point data is generated based on the radian rule of the palm print main line.

[0041] Here, the main line adjustment point corresponding to the adjustment point data is used to control the arc degree of the main line formed by the first main line positioning point and the second main line positioning point.

[0042] The palm print main line radii rule is used to indicate the radii of the main lines in a palm print. For example, the palm print main line is usually not a straight line segment but a curve with a certain radii. After determining the first and second main line positioning points, the radii of the main line formed by the first and second main line positioning points are adjusted by the generated main line adjustment points.

[0043] In one alternative embodiment, the adjustment point data is determined in the area between the first main line positioning point and the second main line positioning point according to the radian rule of the palm print main line.

[0044] Illustratively, Bezier curves are used to parameterize and describe the geometric appearance of the palm print. Optionally, at least one Bezier curve is used to describe the main lines of the palm print of the palm.

[0045] A Bezier curve is a mathematical curve used in two-dimensional graphics applications. A Bezier curve is composed of line segments and nodes. The nodes are draggable fulcrums, and the line segments are like elastic rubber bands. The shape of the line segments can be controlled by controlling the radians of the line segments with the nodes, resulting in a corresponding curve.

[0046] Optionally, by taking any one palm print main line as an example, a quadratic Bezier curve is used to construct the palm print main line, that is, one palm print main line (Bezier curve) is determined using three data (parameter points) in a two-dimensional (2D) plane, where the three data are first data representing a first main line positioning point, a main line adjustment point, and second data representing a second main line positioning point.

[0047] In one alternative embodiment, a palm print main line is generated based on the first data, the second data, and the adjustment point data, where the palm print main line is a curve that sequentially connects the first main line positioning point, the main line adjustment point, and the second main line positioning point.

[0048] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0049] In step 220, a palm print distribution region where at least one palm print main line is located is determined, and at least one palm print thin line is generated within the palm print distribution region according to a predetermined palm print thin line distribution rule.

[0050] For example, after generating at least one palmprint main line, a palmprint distribution area in which the at least one palmprint main line is located is determined, and the palmprint distribution area may be preset and include at least one palmprint main line. Specifically, in the plane in which the at least one palmprint main line is located, a palmprint distribution area that completely includes the at least one palmprint main line can be determined based on the shape of the predetermined palmprint distribution area. Among the at least one palmprint main line, there may be a palmprint main line that connects to the boundary of the palmprint distribution area, or the distance between each of the at least one palmprint main line and the boundary of the palmprint distribution area may not exceed a predetermined distance.

[0051] For example, after generating one palmprint main line, the palmprint main line is treated as a diagonal line, and a rectangular area that can contain the palmprint main line is obtained as the palmprint distribution area; alternatively, after generating multiple palmprint main lines, an area that can contain multiple palmprint main lines is obtained as the palmprint distribution area, and the area may be a regular rectangular area or an irregular area.

[0052] In some embodiments, one palmprint generation area is preset, and at least one palmprint main line is generated within the palmprint generation area, making the palmprint generation area a palmprint distribution area.

[0053] Here, the palm print fine line distribution rule is used to indicate the distribution of fine lines in a palm print.

[0054] Optionally, by analyzing the palm prints of many palms owned by the living being, it is determined that the palm prints of the palms include not only main lines but also fine lines, for example, the distribution of the fine lines is relatively random compared to the main lines, which have a relatively regular distribution, and the fine lines have thinner, shorter and shallower line features compared to the main lines, which have longer, thicker and deeper line features.

[0055] At least one palm print thin line is generated within the palm print distribution area based on the palm print thin line distribution rule.

[0056] Optionally, the palm print thin line distribution rule includes various distribution rules, such as a palm print thin line length rule, a palm print thin line thickness rule, and a palm print thin line density rule.

[0057] Illustratively, the palm print thin line length rule may be used to indicate a palm print thin line length restriction, such as, for example, the length of the palm print thin line being shorter than the shortest palm print main line among multiple palm print main lines, or the length of the palm print thin line being shorter than a predetermined length threshold (e.g., 3 centimeters).

[0058] Illustratively, the palm print thin line thickness rule may be used to indicate a palm print thin line thickness restriction, such as, for example, that the thickness of the palm print thin line is thinner than the thinnest palm print main line among multiple palm print main lines, or that the thickness of the palm print thin line is thinner than a predetermined thickness threshold (e.g., 1 millimeter).

[0059] For example, the palm print line density rule may be used to indicate the mutual distribution of at least two palm print lines within the palm print generation area, such as by specifying that at least three palm print lines are to be generated in a predetermined X region (predetermined region) within the palm print generation area, or by dividing the palm print generation area into a number of sub-regions of unit length (1 centimeter) and specifying that each sub-region contains at least two palm print lines.

[0060] It should be noted that the above various distribution rules may be applied alone or in combination, for example, palm print thin lines may be determined using only the length rule, or palm print thin lines may be determined by comprehensively considering the length rule, thickness rule, and density rule, etc. The above content is an exemplary list, and the embodiments of the present application are not limited thereto.

[0061] In step 230, a candidate palm print image is generated based on at least one palm print main line and at least one palm print thin line in the palm print distribution region, and the candidate palm print image is labeled with a candidate image label, which is used to indicate the palm print distribution of the candidate palm print image, and the palm print distribution includes at least one of a palm print main line distribution and a palm print thin line distribution.

[0062] Here, the palm print distribution includes at least one of a palm print main line distribution and a palm print fine line distribution.

[0063] For example, after obtaining at least one palm print main line and at least one palm print thin line, a palm print distribution region including at least one palm print main line and at least one palm print thin line is selected as a candidate palm print image. In some embodiments, the palm print distribution region including at least one palm print main line and at least one palm print thin line can be further processed to generate a candidate palm print image, such as by extending the palm print distribution region outward from the edge of the palm print distribution region by a predetermined width or adding noise.

[0064] For example, as shown in FIG. 4, three palmprint main lines 410 (thick, long line segments in FIG. 4) are generated based on a predetermined palmprint main line distribution rule, and in the palmprint distribution region where the four palmprint main lines are located, twelve palmprint thin lines 410 (thin, short line segments in FIG. 4) are generated based on a predetermined palmprint thin line distribution rule. The region including the four palmprint main lines and the twelve palmprint thin lines is determined as a candidate palmprint image; that is, the image shown in FIG. 4 is determined as a candidate palmprint image.

[0065] Here, the candidate palm print images are labeled with candidate image labels, which are used to indicate the palm print distribution of the candidate palm print images.

[0066] Here, the palmprint distribution includes at least one of a palmprint main line distribution and a palmprint fine line distribution. The palmprint main line distribution may include description information for each palmprint main line and the relative positional relationship of multiple palmprint main lines. The description information for each palmprint main line may include position description information describing the position of the palmprint main line or direction description information describing the direction of the palmprint main line (the tendency of the main lines in the palmprint to extend). The palmprint fine line distribution may include description information for each palmprint fine line and the relative positional relationship of multiple palmprint fine lines. The description information for each palmprint fine line may include position description information describing the position of the palmprint fine line or direction description information describing the direction of the palmprint fine line (the tendency of the fine lines in the palmprint to extend).

[0067] In some embodiments, after generating a candidate palmprint image, the palmprint main line distribution in the candidate palmprint image is determined, and based on the palmprint main line distribution, a candidate image label corresponding to the candidate palmprint image is determined. Here, the palmprint main line distribution includes at least one of the following situations.

[0068] (1) Determine the palmprint main line distribution based on the relative positional relationship between the first main line positioning point and the second main line positioning point within the palmprint main line.

[0069] Exemplarily, after obtaining at least one palmprint main line, determine the first main line positioning point and the second main line positioning point corresponding to each palmprint main line. When determining the distribution of the target palmprint main line among at least one palmprint main line, determine the first position in the palmprint distribution area of the first main line positioning point within the target palmprint main line, determine the second position in the palmprint distribution area of the second main line positioning point within the target palmprint main line, and determine the relative positional relationship between the first position and the second position in the palmprint distribution area, thereby obtaining the relative positional relationship corresponding to the target palmprint main line. Based on this method, determine the relative positional relationships corresponding to at least one palmprint main line.

[0070] Optionally, comprehensively consider the relative positional relationships corresponding to at least one palmprint main line to determine the palmprint main line distribution corresponding to the candidate palmprint image.

[0071] (2) Determine the palmprint main line distribution based on the relative main line positional relationship between multiple palmprint main lines.

[0072] Exemplarily, when the obtained candidate palmprint image includes multiple palmprint main lines, determine the corresponding main line positions in the palmprint distribution areas of the multiple palmprint main lines to obtain multiple main line positions. Determine the relative positional relationship between the multiple main line positions within the palmprint distribution area. For example, the main line position A corresponding to palmprint main line 1 and the main line position B corresponding to palmprint main line 2 tend to be parallel, and the main line position A corresponding to palmprint main line 1 and the main line position C corresponding to palmprint main line 3 tend to be in a "person" shape, etc.

[0073] Optionally, the relative positional relationships between the plurality of main line positions are comprehensively taken into consideration to determine the palm print main line distribution corresponding to the candidate palm print image.

[0074] In some embodiments, after generating the candidate palm print images, a palm print fine line distribution in the candidate palm print images is determined, and a candidate image label corresponding to the candidate palm print image is determined based on the palm print fine line distribution.

[0075] For example, the palm print fine line distribution is determined based on the relative positional relationship between a plurality of palm print fine lines, or the palm print fine line distribution is determined based on the intersecting relationship between a plurality of palm print fine lines.

[0076] In some embodiments, after generating a candidate palm print image, a distribution of palm print main lines and palm print fine lines in the candidate palm print image is determined, and a candidate image label corresponding to the candidate palm print image is determined based on the distribution of palm print main lines and palm print fine lines.

[0077] Illustratively, the palm print distribution corresponding to the candidate palm print image is used to indicate the identity document (ID) that the candidate palm print image represents, i.e., the candidate image label of the palm print distribution indicates that multiple candidate palm print images can be distinguished.

[0078] Optionally, after generating a plurality of candidate palmprint images, candidate image labels are assigned to the candidate palmprint images by comprehensively considering the distribution of palmprint main lines and palmprint thin lines. If any two candidate palmprint images have a difference in any one of the contents such as the position or length of the palmprint main lines or palmprint thin lines, the candidate image labels will be different, thereby enabling the plurality of candidate palmprint images to be distinguished.

[0079] In some embodiments, a distribution variation range is preset, including a distribution variation range of palmprint main lines and a distribution variation range of palmprint fine lines. If the palmprint main line distribution exceeds the distribution variation range of palmprint main lines, a different candidate image label is assigned; if the palmprint fine line distribution exceeds the distribution variation range of palmprint fine lines, a different candidate image label is assigned; or if the palmprint main line distribution exceeds the distribution variation range of palmprint main lines and the palmprint fine line distribution exceeds the distribution variation range of palmprint fine lines, a different candidate image label is assigned.

[0080] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0081] In step 240, the candidate palm print image is input into a pre-trained palm print generation model to obtain a sample palm print image.

[0082] Here, the palmprint generation model is a model that is pre-trained using palmprint image labels and candidate image labels of palmprint images collected in advance.

[0083] The palm print image is used to represent an image having texture characteristics of a palm print. Optionally, the palm print image is a single image collected in advance, or the palm print image is a plurality of images collected in advance, etc.

[0084] Optionally, the palm print images are labeled with palm print image labels, which are used to indicate the palm print distribution of the palm print images.

[0085] Here, the palm print distribution includes at least one of a palm print main line distribution and a palm print fine line distribution, and the palm print distribution corresponding to the palm print image is used to indicate the identification information indicated by the palm print image.

[0086] Optionally, the palm print main lines are determined by at least the palm print main line start point, the palm print main line end point, and the arc degree of the palm print main lines, and at least one palm print main line distribution is used as a palm print image label corresponding to the palm print image.

[0087] For example, if a palm print image has multiple palm print main lines, the palm print image label corresponding to the palm print image is determined based on the starting point, ending point, and arc degree of the longest palm print main line. Alternatively, if a palm print image has multiple palm print main lines, the palm print image label corresponding to the palm print image is determined based on the starting point, ending point, and arc degree of the palm print main lines, respectively, corresponding to the multiple palm print main lines.

[0088] Optionally, at least one palm print fine line distribution is used as a palm print image label corresponding to the palm print image.

[0089] For example, the palm print image labels are used not only to indicate the distribution of the palm print main lines, but also to indicate the distribution of multiple palm print thin lines in the palm print image, such as the positional relationship of palm print thin lines in the palm print image or the quantity status of palm print thin lines in the palm print image.

[0090] Optionally, at least one palm print fine line distribution and at least one palm print main line distribution are used as a palm print image label corresponding to the palm print image, etc.

[0091] Here, the sample palm print images are used to train a palm print recognition model, and the palm print recognition model is used to perform palm print recognition.

[0092] Illustratively, after adjusting the candidate palm print images, a sample palm print image is obtained that is more consistent with the palm print distribution rules in the collected palm print images.

[0093] In some embodiments, the sample palmprint image is used to train a palmprint recognition model to be trained; or the sample palmprint image is mixed with multiple palmprint images in a palmprint image library, and then the palmprint recognition model to be trained is trained through multiple images (including the sample palmprint image and multiple palmprint images); or the sample palmprint image is used to initially train the palmprint recognition model to be trained, and then the multiple palmprint images in the palmprint image library are used to train the initially trained palmprint recognition model, thereby improving the robustness of the palmprint recognition model.

[0094] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0095] In summary, since the candidate palmprint images are images obtained based on the palmprint main line distribution rules and palmprint fine line distribution rules, they have the characteristics of being large in quantity and diverse. At the same time, in order to make the candidate palmprint images closer to real palmprint images, a palmprint generation model is pre-trained using the palmprint image labels and candidate image labels of pre-collected palmprint images, and the palmprint generation model is used to generate sample palmprint images corresponding to the candidate palmprint images. This makes the palmprint distribution in the sample palmprint images more accurate, not only breaking through the limitation of a small number of palmprints, but also improving the reliability of the sample palmprint images, thereby improving the robustness of the palmprint recognition model and helping to improve the palmprint recognition accuracy of the palmprint recognition model.

[0096] In one alternative embodiment, the palmprint generation model is a model obtained by performing model training on a candidate generative model. Illustratively, as shown in Fig. 5, the process of obtaining a palmprint generation model by performing model training on a candidate generative model can be realized as the following steps 510 to 550.

[0097] In step 510, a palm print image library is obtained.

[0098] Here, the palm print image library stores a plurality of collected palm print images in advance, and each of the plurality of palm print images is assigned a palm print image label.

[0099] For example, the plurality of palmprint images are images having palmprint texture features that have been collected in advance. For example, a palmprint image library stores palmprint images a1 and a2 of object A, for example, the palmprint image a1 of object A is a palmprint image of the left hand of object A, and the palmprint image a2 of object A is a palmprint image of the right hand of object A. The palmprint image library further stores palmprint images b1 and b2 of object B, for example, the palmprint image b1 of object B is a palmprint image of the left hand of object B, and the palmprint image b2 of object B is another image of the palmprint of the left hand of object B.

[0100] Here, a palm print image label is attached to each of the plurality of palm print images, and the palm print image label is used to indicate the palm print distribution of the palm print images.

[0101] In step 520, features are extracted from the palmprint image to obtain an image feature representation corresponding to the palmprint image.

[0102] For example, an image feature representation corresponding to each of the at least one palm print image is obtained by, for example, performing feature extraction on the at least one palm print image by an encoder, and obtaining an image feature representation corresponding to each of the at least one palm print image.

[0103] In step 530, features are extracted from the candidate palmprint image to obtain a candidate feature representation corresponding to the candidate palmprint image.

[0104] For example, after obtaining a candidate palmprint image, a downsampling operation is performed on the candidate palmprint image to obtain a candidate feature representation corresponding to the candidate palmprint image, or feature extraction is performed on the candidate palmprint image by an encoder to obtain a candidate feature representation corresponding to the candidate palmprint image, etc.

[0105] In some embodiments, a candidate generative model is trained to perform feature extraction on the candidate palm print image to obtain candidate feature representations corresponding to the candidate palm print image.

[0106] In step 540, the candidate generative model to be trained fuses and upsamples the candidate feature representation and the image feature representation to obtain a candidate sample palmprint image.

[0107] In some embodiments, the candidate generative model performs a feature fusion process on the candidate feature representation and the image feature representation to obtain a fused feature representation.

[0108] Here, the fused feature representation is used to represent a feature representation obtained by performing constraint adjustment on a candidate feature representation using at least one image feature representation.

[0109] For example, a fused feature representation is obtained by performing a position-wise dot product operation on the candidate feature representation and the image feature representation. For example, if the feature dimension of the image feature representation is 64 and the feature dimension of the candidate feature representation is also 64, the feature dimension of the fused feature representation will be 64.

[0110] Optionally, a feature fusion process is performed on the candidate feature representations via the at least one image feature representation to obtain fused feature representations respectively corresponding to the at least one image feature representation, i.e., to obtain at least one fused feature representation.

[0111] Optionally, a feature fusion process is performed on the multiple image feature representations and the candidate feature representations to obtain one fusion feature representation, which is used to represent the result of adjusting the candidate feature representation with the multiple image feature representations.

[0112] In some embodiments, an upsampling process is performed on the fused feature representation to obtain candidate sample palmprint images.

[0113] Optionally, after obtaining the fused feature representations, a size restoration is performed on the at least one fused feature representation by an upsampling process to obtain sample palmprint images respectively corresponding to the at least one fused feature representations.

[0114] In step 550, the candidate sample palm print images are used to train the candidate generative model to obtain a palm print generative model.

[0115] In one alternative embodiment, a candidate palm print image corresponding to the candidate sample palm print image is determined, and the candidate image label corresponding to the candidate palm print image is used as the sample image label of the candidate sample palm print image.

[0116] For example, since a candidate sample palm-print image is an image obtained based on a candidate feature representation and an image feature representation, the candidate sample palm-print image corresponds to the candidate palm-print image corresponding to the candidate feature representation. For example, since candidate sample palm-print image A is an image obtained based on candidate feature representation a1 and image feature representation b1, the candidate sample palm-print image corresponds to candidate palm-print image A1 corresponding to candidate feature representation a1.

[0117] Optionally, based on the correspondence between the candidate sample palm print image and the candidate palm print image, a candidate image label of the candidate palm print image corresponding to the candidate sample palm print image is determined, and the candidate image label is used as the sample image label of the candidate sample palm print image.

[0118] For example, if the candidate image label of candidate palm-print image A1 corresponding to candidate sample palm-print image A is "label a," then "label a" is used as the sample image label of the candidate sample palm-print image.

[0119] In one alternative embodiment, a loss value corresponding to the candidate sample palmprint image is obtained based on the distribution difference between the palmprint distribution indicated by the sample image label and the palmprint distribution indicated by the palmprint image label.

[0120] Illustratively, a loss value corresponding to each of the sample image label and the at least one palmprint image label is calculated.

[0121] For example, a first image feature representation corresponding to a sample image label and a second image feature representation corresponding to each of at least one palmprint image are obtained, and a loss value corresponding to each of at least one palmprint image label is obtained based on the vector distance between the first image feature representation and the second image feature representation in the vector space.

[0122] In one alternative embodiment, the loss value is used to train the candidate generative model to obtain a palmprint generative model.

[0123] Illustratively, the candidate generative model is trained using at least one loss value until the training of the candidate generative model reaches a training condition, thereby obtaining a palm print generative model.

[0124] Here, the palm print generation model is used to generate a sample palm print image, i.e., the palm print generation model is used in the process of generating a sample palm print image.

[0125] In summary, a pre-trained palmprint generation model is obtained using the palmprint image labels and candidate image labels of pre-collected palmprint images, and the palmprint generation model is used to generate sample palmprint images corresponding to the candidate palmprint images, thereby making the palmprint distribution in the sample palmprint images more accurate, not only breaking through the limitation of a small number of palmprints, but also improving the reliability of the sample palmprint images, thereby improving the robustness of the palmprint recognition model and helping to improve the palmprint recognition accuracy of the palmprint recognition model.

[0126] In this embodiment, we introduce a model training process for training a candidate generative model to obtain a palmprint generation model for generating a sample palmprint image. The image feature representation corresponding to a pre-acquired palmprint image and the candidate feature representation of the candidate palmprint image are combined and upsampled, and at least one image feature representation is used to constrain and adjust the candidate feature representation to obtain a candidate sample palmprint image. Compared with a candidate palmprint image that has not been adjusted, the candidate sample palmprint image is closer to the palmprint texture features of the pre-acquired palmprint image. Training the candidate generative model using the candidate sample palmprint image provides a more robust palmprint generation model, which is useful for generating more accurate and realistic sample palmprint images.

[0127] In one alternative embodiment, a sample palmprint image can be directly obtained based on a pre-trained palmprint generation model, or a sample palmprint image can be obtained by using the palmprint image to perform constraint adjustment on the candidate palmprint image by utilizing the label association relationship between the collected palmprint images and the image labels corresponding to the generated candidate palmprint images. For example, the collected palmprint images and the generated candidate palmprint images can be input together into the palmprint generation model, which can be implemented as the following steps 610 to 640, as shown in FIG. 6 .

[0128] In step 610, at least one palm print main line is generated based on a predetermined palm print main line distribution rule.

[0129] In some embodiments, the palm print can be realized as a palm print of a hand, in which the distribution of the palm print main lines usually exhibits a diagonal relationship, that is, the distribution rule of the palm print main lines is a diagonal rule.

[0130] In some embodiments, the palm print can further be realized as a sole print. The distribution of the main lines of the palm print in the sole print usually appears in various palm print forms such as various "V"-shaped lines, tortoise shell-shaped lines, vertical lines, etc. That is, the distribution rule of the main lines of the palm print on the sole can be determined in comparison with some common forms.

[0131] In one alternative embodiment, positioning point data is generated based on a predetermined distribution rule of the main lines of the palm print.

[0132] Here, the positioning point data includes first data corresponding to the first main line positioning point and second data corresponding to the second main line positioning point.

[0133] In one alternative embodiment, the starting point of the main line of the palm print is taken as the first main line positioning point, and the ending point of the main line of the palm print is taken as the second main line positioning point. Exemplarily, the generation of the positioning point data includes at least one of the following methods.

[0134] (1) Random generation method Exemplarily, in any one coordinate region, the first data and the second data are generated in a random generation manner. Based on the length limit of the main line of the palm print, the line segment connecting the first main line positioning point and the second main line positioning point is scaled in the same ratio as the coordinate region to determine the first main line positioning point corresponding to the first data and the second main line positioning point corresponding to the second data.

[0135] (2) Generation method within a divided region Optionally, taking the pre-set distribution region of the main lines of the palm print as an example, in the distribution region of the main lines of the palm print, the region is divided to obtain divided regions, and in the divided regions, the first data and the second data are generated in a random generation manner. Exemplarily, after the region is divided for the distribution region, two divided regions are obtained, the first main line positioning point corresponding to the first data is generated in the first divided region, and the second main line positioning point corresponding to the second data is generated in the second divided region.

[0136] Optionally, when the distribution area is divided into areas, multiple divided areas are obtained, and the sum of the areas of the multiple divided areas may be realized as all of the distribution area or as part of the distribution area, i.e., the divided area for generating the positioning point data may include all of the distribution area or only part of the distribution area, etc.

[0137] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0138] In one alternative embodiment, the adjustment point data is generated based on the radian rule of the palm print main line.

[0139] Here, the main line adjustment point corresponding to the adjustment point data is used to control the arc degree of the main line formed by the first main line positioning point and the second main line positioning point.

[0140] In one alternative embodiment, the adjustment point data is determined in the area between the first main line positioning point and the second main line positioning point according to the radian rule of the palm print main line.

[0141] 7 is a schematic diagram of determining palmprint main lines using the Bezier curve method. The area enclosed by the horizontal and vertical axes is the palmprint main line generation area, and the numbers marked on the horizontal and vertical axes are used to assist in determining the coordinate positions of three data. The diagram includes three palmprint main lines, each determined by first data, main line adjustment points, and second data. The "inverted triangle symbol" indicates the first main line positioning point corresponding to the first data, the "star symbol" indicates the main line adjustment point, and the "circle symbol" indicates the second main line positioning point corresponding to the second data.

[0142] For example, first, a first main line positioning point corresponding to the first data and a second main line positioning point corresponding to the second data are generated, and then a main line adjustment point is generated between the first main line positioning point and the second main line positioning point. Optionally, the main line adjustment point is generated based on a coordinate region restriction relationship between the first main line positioning point and the second main line positioning point.

[0143] For example, the coordinates of the first main line positioning point of the palm print main line 710 are (0.0,0.4), the coordinates of the main line adjustment point are (0.5,0.6), and the coordinates of the second main line positioning point are (0.6,1.0). The coordinates of the first main line positioning point of the palm print main line 720 are (0.0,0.0), the coordinates of the main line adjustment point are (0.5,0.3), and the coordinates of the second main line positioning point are (1.0,1.0). The coordinates of the first main line positioning point of the palm print main line 730 are (0.3,0.0), the coordinates of the main line adjustment point are (0.6,0.3), and the coordinates of the second main line positioning point are (1.0,0.3).

[0144] In one alternative embodiment, at least one palm print main line is generated based on the first data, the second data, and the adjustment point data.

[0145] Here, the palm print main line is a curve that sequentially connects the first main line positioning point, the main line adjustment point, and the second main line positioning point.

[0146] For example, as shown in FIG. 7 , the palm print main line 710 is obtained by sequentially connecting the first main line positioning point, the main line adjustment point, and the second main line positioning point, and is an arc-shaped curve that protrudes slightly upward; the palm print main line 720 is obtained by sequentially connecting the first main line positioning point, the main line adjustment point, and the second main line positioning point, and is an arc-shaped curve that protrudes slightly upward; and the palm print main line 730 is obtained by sequentially connecting the first main line positioning point, the main line adjustment point, and the second main line positioning point, and is an arc-shaped curve that concaves downward.

[0147] In step 620, a palm print distribution region where at least one palm print main line is located is determined, and at least one palm print thin line is generated within the palm print distribution region according to a predetermined palm print thin line distribution rule.

[0148] In some embodiments, after generating at least one palmprint main line, a palmprint distribution area in which the palmprint main line is located is determined, or after pre-determining the palmprint distribution area in which the palmprint main line is located, at least one palmprint main line is generated in the palmprint distribution area, which becomes the palmprint distribution area in which the at least one palmprint main line is located.

[0149] In one alternative embodiment, after generating at least one palmprint main line in the palmprint distribution area, at least one palmprint thin line is generated based on a predetermined palmprint thin line distribution rule; alternatively, simultaneously generating at least one palmprint main line in the palmprint distribution area based on the palmprint main line distribution rule and generating at least one palmprint thin line based on the predetermined palmprint thin line distribution rule, etc.

[0150] Here, the palm print fine line distribution rule is used to indicate the distribution of fine lines in a palm print.

[0151] For example, after generating three palmprint main lines within the palmprint distribution area, 13 palmprint thin lines are generated based on a predetermined palmprint thin line distribution rule, or the palmprint main lines and palmprint thin lines are generated based on the palmprint main line distribution rule and the palmprint thin line distribution rule, resulting in a total of four palmprint main lines and 22 palmprint thin lines.

[0152] Optionally, the main palm print lines and the thin palm print lines may intersect, the main palm print lines may intersect with each other, and the thin palm print lines may intersect with each other.

[0153] In one alternative embodiment, within the palm print distribution region where at least one palm print main line is located, positioning points of a plurality of palm print thin lines are determined based on palm print thin line distribution rules.

[0154] Here, the positioning points of the palm print thin lines are used to determine the thin line generation range, and exemplarily, at least two palm print thin line positioning points are generated based on the palm print thin line distribution rule.

[0155] Optionally, the palm print thin line distribution rule includes various distribution rules such as a length rule, a thickness rule, and a density rule of the palm print thin line, and the various distribution rules may be applied alone or in combination. For example, the palm print thin line may be determined using only the length rule, or the palm print thin line may be determined by comprehensively considering the length rule, thickness rule, and density rule.

[0156] Optionally, in the palm print distribution, the distribution of palm print thin lines is relatively dispersed and random, and based on the palm print thin line distribution rule and the palm print distribution of the palm print data stored in the palm print database, at least one of the following methods is used to generate at least one palm print thin line in the palm print generation area:

[0157] (1) After determining the positioning point of at least one palm print thin line, determine the positioning point of at least one remaining palm print thin line based on the palm print thin line distribution rule.

[0158] For example, the predetermined length threshold of the palm print thin line is set to 3 centimeters, and the predetermined thickness threshold of the palm print thin line is set to 1 millimeter. In the palm print generation area, a positioning point of one palm print thin line is randomly generated, and at least one positioning point of the remaining palm print thin line is determined within the predetermined length threshold and the predetermined thickness threshold of the palm print thin line, thereby obtaining positioning points of at least two palm print thin lines. Optionally, based on the positioning point of the at least one remaining palm print thin line, positioning points of the remaining palm print thin line are further determined within the predetermined length threshold and the predetermined thickness threshold of the palm print thin line.

[0159] (2) Randomly determine the positioning points of at least two palm print fine lines.

[0160] For example, in the palm print generation area, at least two points are randomly generated as positioning points of at least two palm print thin lines based on the palm print thin line distribution rule, or after dividing the palm print generation area, at least two sub-areas are obtained, and the positioning points of the palm print thin lines for generating palm print thin lines in the sub-areas are determined.

[0161] In one alternative embodiment, at least one palm print fine line is generated based on the positioning points of the plurality of palm print fine lines.

[0162] Optionally, after obtaining the positioning points of at least two palm print thin lines, the palm print thin line obtaining manner includes at least one of the following:

[0163] 1. At least two of the positioning points of the plurality of palm print thin lines are connected to obtain at least one palm print thin line.

[0164] For example, after obtaining the positioning points of the generated multiple palm print thin lines, the positioning points of any two palm print thin lines can be connected to obtain a palm print thin line in the form of a line segment, or the positioning points of any multiple palm print thin lines can be connected to obtain a palm print thin line in the form of an irregular line segment, or taking into account the condition that the length of the thin line is short, the positioning points of any multiple (two or more) palm print thin lines can be taken as a set of thin line positioning points to obtain a palm print thin line within a certain length range.

[0165] 2. Determining at least one palm print sample based on palm print fine line adjustment points generated by the positioning points of the plurality of palm print fine lines.

[0166] In one alternative embodiment, a palm print fine line adjustment point is determined based on the positioning points of at least two palm print fine lines.

[0167] Here, the palm print thin line adjustment point is used to control the arc degree between at least two palm print thin line positioning points among the plurality of palm print thin line positioning points.

[0168] For example, when generating palm print thin lines, the palm print thin lines are determined using the Bezier curve method, that is, after generating at least two palm print thin line positioning points, a palm print thin line adjustment point is determined to adjust the arc degree of the line segment between the at least two palm print thin line positioning points, and in the process of moving the position of the palm print thin line adjustment point, different palm print thin lines are obtained when the palm print thin line adjustment point is located at a different position, for example, a palm print thin line represented as an arc line with a large arc degree or represented as an irregular curve with a small curvature width.

[0169] In one alternative embodiment, the palm print fine lines are determined based on at least two palm print fine line positioning points and palm print fine line adjustment points within a preset palm print fine line quantity range.

[0170] Optionally, the number of palmprint main lines and palmprint thin lines is preset to make the palmprint situation in the generated palmprint sample similar to the palmprint situation of a living organism. For example, take the observation results of a certain palm area as an example. In the palmprint situation of a living organism, the number of palmprint main lines within the palm area is 2 to 5, and the number of palmprint thin lines is 5 to 15. For example, the number of palmprint main lines is preset to 2 to 5, and the number of palmprint thin lines is preset to 5 to 15. When determining the palmprint main lines, the number of palmprint main lines is controlled within the range of 2 to 5 to obtain a maximum of four different situations of palmprint main lines, and when determining the palmprint thin lines, the number of palmprint thin lines is controlled within the range of 5 to 15 to obtain a maximum of 11 different situations of palmprint thin lines.

[0171] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0172] In step 630, a candidate palm print image is generated based on at least one palm print main line and at least one palm print thin line in the palm print distribution region, and the candidate palm print image is labeled with a candidate image label, which is used to indicate the palm print distribution of the candidate palm print image, and the palm print distribution includes at least one of a palm print main line distribution and a palm print thin line distribution.

[0173] In one alternative embodiment, candidate palm print images are generated that include palm print main lines and palm print thin lines within a preset palm print quantity range.

[0174] Wherein, the palmprint quantity range includes at least one of a palmprint main line quantity range and a palmprint fine line quantity range. Optionally, the candidate palmprint image includes a certain quantity of palmprint main lines and palmprint fine lines, where when a palmprint sample is obtained based on the palmprint situation of a living organism, the quantity of palmprint main lines and palmprint fine lines in the palmprint sample is similar to the quantity of palmprint main lines and palmprint fine lines of the living organism.

[0175] For example, in a certain palm area, the number of palmprint main lines is controlled to 2 to 5, and the number of palmprint thin lines is controlled to 5 to 15, to obtain multiple candidate palmprint images. For example, the first candidate palmprint image has 3 main lines and 12 thin lines, the second candidate palmprint image has 5 main lines and 10 thin lines, and the third candidate palmprint image has 4 main lines and 5 thin lines.

[0176] Here, the candidate palm print image is labeled with a candidate image label, and the candidate image label is used to indicate the palm print distribution of the candidate palm print image, where the palm print distribution includes at least one of a palm print main line distribution and a palm print fine line distribution.

[0177] In one alternative embodiment, a candidate image label corresponding to the candidate palmprint image is determined based on the relative positional relationship between the first data and the second data in the candidate palmprint image.

[0178] For example, the first data indicates data corresponding to the starting points of palmprint main lines, and the second data indicates data corresponding to the ending points of palmprint main lines. When determining a candidate image label corresponding to a candidate palmprint image, a plurality of palmprint main lines in the candidate palmprint image are determined, and palmprint main line starting points and palmprint main line ending points corresponding to the plurality of palmprint main lines are determined. Based on the relative positional relationships between the palmprint main line starting points and palmprint main line ending points corresponding to each palmprint main line, relative positional information corresponding to each palmprint main line is determined, and the relative positional information corresponding to the plurality of palmprint main lines is comprehensively taken into consideration to obtain a candidate image label corresponding to the candidate palmprint image.

[0179] In one alternative embodiment, a candidate image label corresponding to the candidate palm print image is determined based on the relative positional relationship of the positioning points of the multiple palm print thin lines in the candidate palm print image, or a candidate image label corresponding to the candidate palm print image is determined based on the positioning points of the multiple palm print thin lines and the palm print thin line adjustment points.

[0180] Here, the candidate image labels assigned to the candidate palmprint images are used to represent the identification information indicated by the candidate palmprint images. For example, candidate palmprint image 1 is assigned candidate image label a1 based on the palmprint main line distribution and palmprint fine line distribution, and candidate palmprint image 2 is assigned candidate image label a2 based on the palmprint main line distribution and palmprint fine line distribution.

[0181] In step 640, the candidate palmprint image and the palmprint image are input into a pre-trained palmprint generation model, and the palmprint image is used to perform constraint adjustment on the candidate palmprint image to obtain a sample palmprint image.

[0182] In one alternative embodiment, features are extracted from the candidate palmprint image to obtain a candidate feature representation corresponding to the candidate palmprint image.

[0183] For example, after obtaining a candidate palmprint image consisting of generated palmprint main lines and generated palmprint thin lines, feature extraction is performed on the candidate palmprint image to obtain a candidate feature representation corresponding to the candidate palmprint image, and the candidate feature representation indicates image information of the generated candidate palmprint image.

[0184] In some embodiments, the candidate feature representation corresponding to the candidate palmprint image is obtained by performing a downsampling operation on the candidate palmprint image, or by passing the candidate palmprint image through a general feature extraction model.

[0185] In one alternative embodiment, feature extraction is performed on the palm print images to obtain image feature representations corresponding to at least one palm print image, respectively.

[0186] Illustratively, in addition to performing feature extraction on the candidate palm print image, feature extraction is performed on at least one palm print image in a palm print image library, where the at least one palm print image may be realized as a plurality of palm print images in the palm print image library or as at least one palm print image arbitrarily selected from the plurality of palm print images.

[0187] Optionally, at least one palm print image is input to an encoder to extract image feature representations corresponding to the at least one palm print image respectively, and the encoder is configured by a three-layer convolutional network.

[0188] In one alternative embodiment, at least one image feature representation is used to perform constraint adjustment on the candidate feature representation based on the palmprint distribution difference conditions indicated by the label association relationship to obtain a sample palmprint image.

[0189] Illustratively, the candidate image label indicates the palm print distribution of the candidate palm print image, and the palm print image label indicates the palm print distribution of the palm print image. Optionally, the label association relationship between the candidate image label and the palm print image label is realized as a label difference relationship, and the label difference relationship is used to indicate a difference situation of the palm print distribution between the generated candidate palm print image and the collected palm print image.

[0190] In one alternative embodiment, palm print image labels each corresponding to the at least one image feature representation are determined.

[0191] Illustratively, after extracting image feature representations respectively corresponding to at least one palm print image, palm print image labels respectively corresponding to the at least one image feature representation are determined.

[0192] In one alternative embodiment, feature fusion is performed on at least one image feature representation and candidate feature representations based on the difference in palmprint distribution indicated by the label association relationship, to obtain image fusion feature representations corresponding to the at least one image feature representation.

[0193] In some embodiments, an upsampling process is performed on the at least one image fusion feature representation to obtain sample palm print images respectively corresponding to the at least one image fusion feature representation.

[0194] Optionally, after obtaining the fusion feature representation, a size restoration is performed on the at least one image fusion feature representation by an upsampling process to obtain sample palmprint images respectively corresponding to the at least one image fusion feature representation.

[0195] The image fusion feature representation is obtained by adjusting the candidate feature representation with the palm print feature representation corresponding to the collected palm print image, so that it can better reflect the reliability of the palm print, and the sample palm print image obtained based on the fusion feature representation can also better reflect the reliability of the palm print.

[0196] In one preferred embodiment, the feature fusion process for obtaining the above-mentioned image fusion feature representation is realized by a decoder (also called a palmprint generation model). That is, after obtaining a trained decoder, the decoder fuses the image feature representation corresponding to the palmprint image with the candidate feature representation corresponding to the candidate palmprint image, and performs an upsampling process on the image fusion feature representation to obtain a sample palmprint image.

[0197] Here, the sample palm print images are used to train a palm print recognition model, and the palm print recognition model is used to perform palm print recognition.

[0198] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0199] In summary, a pre-trained palmprint generation model is obtained using the palmprint image labels and candidate image labels of pre-collected palmprint images, and the palmprint generation model is used to generate sample palmprint images corresponding to the candidate palmprint images, thereby making the palmprint distribution in the sample palmprint images more accurate, not only breaking through the limitation of a small number of palmprints, but also improving the reliability of the sample palmprint images, thereby improving the robustness of the palmprint recognition model and helping to improve the palmprint recognition accuracy of the palmprint recognition model.

[0200] In the present embodiment, we propose an improved method for generating candidate palmprint images, which involves inputting the candidate palmprint image and the palmprint image into a pre-trained palmprint generation model, and using the palmprint image to perform constrained adjustment on the candidate palmprint image to obtain a sample palmprint image. In this process, in addition to the palmprint generation model obtained through training, the palmprint image is used to perform more targeted adjustment on the candidate palmprint image, so that the palmprint generation model and the palmprint image are taken into consideration comprehensively, thereby improving the reliability of the generation of the sample palmprint image.

[0201] In one alternative embodiment, a palmprint generation model is obtained by training using the loss value determined between the candidate generation model and the palmprint image, and a sample palmprint image is generated. In addition, a feature similarity calculation process is performed using the generated sample palmprint image and the palmprint image, and the feature similarity calculation result is used to constrain the identity consistency (also called identity identity) between the sample palmprint image and at least one palmprint image. For example, as shown in FIG. 8, this can be realized as the following steps 810 to 830.

[0202] In step 810, a target feature extraction model is obtained.

[0203] Here, the target feature extraction model is a feature extraction model that is pre-trained using a plurality of palm print images in a palm print image library, that is, the training process of the target feature extraction model is associated with at least one palm print image.

[0204] In step 820, the target feature extraction model performs feature extraction on the sample palm print image under training and the at least one palm print image, to obtain a first feature representation corresponding to the sample palm print image under training and a second feature representation corresponding to the at least one palm print image.

[0205] Illustratively, after obtaining a sample palmprint image under training and at least one palmprint image, a target feature extraction model is used to perform feature extraction on the sample palmprint image under training to obtain a first feature representation corresponding to the sample palmprint image under training, and the target feature extraction model is used to perform feature extraction on the at least one palmprint image to obtain a second feature representation corresponding to the at least one palmprint image.

[0206] In step 830, label constraints are applied to the sample palm print image based on feature differences between the first feature representation and the at least one second feature representation.

[0207] Here, the label constraint is used to constrain the identity consistency between the sample palm print image and at least one palm print image, i.e., constrain the sample palm print image and the at least one palm print image to show the same identity as much as possible.

[0208] Illustratively, after obtaining a first feature representation representing a sample palmprint image during training and a second feature representation representing the palmprint image, a cosine similarity calculation is performed on the first feature representation and at least one second feature representation to obtain at least one feature similarity.

[0209] In some embodiments, label constraints are applied to the sample palm print images based on at least one feature similarity.

[0210] Optionally, determine identity binding features respectively corresponding to the at least one second feature representation based on the at least one feature similarity, and perform label binding on the sample palmprint image using the identity binding features.

[0211] Illustratively, the cosine similarity sim between any one of the at least one second feature representation and the first feature representation is calculated, and then (1-sim) is used as an identity constraint feature to constrain the identity consistency between the second feature representation and the first feature representation.

[0212] Here, since the first feature representation indicates a sample palmprint image under training, the identity information of the sample palmprint image under training is determined based on the candidate image label attached to the corresponding candidate palmprint image, and when constraining the identity consistency between the second feature representation and the first feature representation, the identity constraint is performed based on the first feature representation.

[0213] Optionally, label constraint is performed on the sample palmprint image corresponding to the first feature representation by reducing a feature distance between the second feature representation and the first feature representation whose feature similarity is greater than a predetermined similarity threshold, and increasing a feature distance between the second feature representation and the first feature representation whose feature similarity is less than the predetermined similarity threshold.

[0214] In some embodiments, the above label constraint method is applied to the training process of a palmprint recognition model to establish an identity association relationship between the generated sample palmprint image and the stored palmprint images, which helps the palmprint recognition model learn more and richer identity information related to the identity association relationship; and when applying a target palmprint recognition model, it helps to determine a palmprint image corresponding to an object to be authenticated from multiple palmprint images stored in a palmprint image library, which helps to improve the authentication accuracy of the target palmprint recognition model.

[0215] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0216] In summary, a pre-trained palmprint generation model is obtained using the palmprint image labels and candidate image labels of pre-collected palmprint images, and the palmprint generation model is used to generate sample palmprint images corresponding to the candidate palmprint images, thereby making the palmprint distribution in the sample palmprint images more accurate, not only breaking through the limitation of a small number of palmprints, but also improving the reliability of the sample palmprint images, thereby improving the robustness of the palmprint recognition model and helping to improve the palmprint recognition accuracy of the palmprint recognition model.

[0217] In the embodiments of this application, a target feature extraction model is used to perform label constraint on a sample palmprint image. Because the target feature extraction model is a feature extraction model pre-trained using a number of palmprint images, the model can strongly focus on the palmprint generation features shown in the palmprint image, improve the label constraint ability of the sample palmprint image, and further establish an identity association relationship between the generated sample palmprint image and the stored palmprint image, which helps ensure the reliability of the sample generated image and avoids the impact on the recognition accuracy of the palmprint recognition model caused by inappropriate training of the palmprint recognition model by unrealistic sample generated images.

[0218] In one alternative embodiment, after obtaining the sample palm print image, the sample palm print image is used to train a palm print recognition model to obtain a trained target palm print recognition model. For example, as shown in Fig. 9, after step 240 shown in Fig. 2 above, the method may further include the following steps 910 to 930.

[0219] In step 910, the sample palm print images are used to train palm print recognition models to obtain candidate palm print recognition models.

[0220] Here, the palmprint recognition model is a model to be trained in advance and is used in the palmprint recognition process to determine palmprint information corresponding to a palmprint image. The palmprint information includes at least one of palmprint distribution information and palmprint identity information, where the palmprint distribution information is used to indicate the distribution of palmprint main lines and / or palmprint thin lines in the palmprint image, and the palmprint identity information is used to indicate object information corresponding to the palmprint image, etc.

[0221] Illustratively, after obtaining sample palm print images, palm print recognition models are trained using the generated sample palm print images to obtain candidate palm print recognition models.

[0222] Optionally, in response to the training of the palm print recognition model reaching a training condition, a candidate palm print recognition model is obtained. For example, when the number of times the palm print recognition model is trained using the generated sample palm print images reaches a predetermined number of training times, the training condition is considered to be met and a candidate palm print recognition model is obtained, or when the training of the palm print recognition model using the generated sample palm print images converges (for example, the loss value remains unchanged during training), the training condition is considered to be met and a candidate palm print recognition model is obtained.

[0223] In step 920, at least one palm print image in the palm print image library is used to train the candidate palm print recognition model to obtain a target palm print recognition model.

[0224] Here, the target palm print recognition model is used to perform palm print recognition.

[0225] Illustratively, after obtaining a candidate palm print recognition model, the candidate palm print recognition model is adjusted using at least one palm print image in the collected palm print image library to improve the recognition accuracy of the candidate palm print recognition model for the real palm print, thereby obtaining a target palm print recognition model with higher recognition accuracy.

[0226] In step 930, the target palm print image is input into the target palm print recognition model to obtain an image authentication result.

[0227] Here, the target palmprint image is used to indicate the palmprint image to be subjected to palmprint recognition. Illustratively, after obtaining the target palmprint recognition model, the target palmprint image to be authenticated is input into the target palmprint recognition model to obtain an image authentication result, which is realized as palmprint information corresponding to the target palmprint image, and includes at least one of palmprint distribution information and palmprint identity information.

[0228] Optionally, take a scenario in which the target palmprint recognition model is applied to hand authentication for implementing a payment process as an example. As shown in FIG. 10 , take an example in which the target palmprint image is an image collected by a terminal payment device 1010. An object's hand image 1020 collected by the terminal payment device 1010 is used as the target palmprint image to be authenticated. When authenticating the object's hand image 1020, in order to accurately authenticate the region, first, region extraction is performed on the object's hand image 1020 to obtain a region of interest 1030 in the object's hand image 1020. For example, a 5 cm*5 cm region of the object's palmprint image 1020 is cropped as the region of interest 1030. The region of interest 1030 includes palmprint main lines and palmprint thin lines, and is the object's palmprint image region that represents the palmprint content in the object's hand image 1020.

[0229] Features are extracted from the region of interest 1030 by the target palm print recognition model to obtain an interest feature representation corresponding to the region of interest 1030. Furthermore, feature extraction is performed on a plurality of palm print images collected in advance to obtain palm print feature representations corresponding to the plurality of palm print images, respectively.

[0230] After obtaining the feature representation of interest and the palm print feature representations respectively corresponding to the plurality of palm print images, cosine similarities respectively corresponding to the feature representation of interest and the plurality of palm print feature representations are calculated to obtain a plurality of cosine similarity results.

[0231] Optionally, the maximum cosine similarity result is selected from the multiple cosine similarity results, and the palm print image corresponding to the maximum cosine similarity result is determined, and the object information corresponding to the palm print image is taken as the calculation result corresponding to the object palm print image 1020. For example, the object palm print image 1020 may be determined to be the object whose identification number (ID: Identity Document) in the multiple palm print images is 36.

[0232] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0233] In summary, the candidate palmprint images are obtained based on the palmprint main line distribution rules and palmprint fine line distribution rules, and therefore have the characteristics of being large in quantity and diverse. At the same time, in order to make the candidate palmprint images closer to real palmprint images, constrained adjustment is performed on the generated candidate palmprint images using pre-collected palmprint images, which makes the palmprint distribution in the adjusted sample palmprint images more accurate, not only breaking through the limitation of a small number of palmprints but also improving the reliability of the sample palmprint images, thereby improving the robustness of the palmprint recognition model and helping to improve the palmprint recognition accuracy of the palmprint recognition model.

[0234] In this embodiment, a training process for training a palmprint recognition model using sample palmprint images is introduced. First, a palmprint recognition model is trained using the generated sample palmprint images to obtain a candidate palmprint recognition model. Then, the candidate palmprint recognition model is adjusted using collected palmprint images to improve the authentication reliability of the candidate palmprint recognition model and obtain a target palmprint recognition model with higher authentication accuracy. As a result, when the target palmprint recognition model authenticates a target palmprint image to be authenticated, more accurate image authentication results can be obtained and the application range of the target palmprint recognition model can be expanded.

[0235] In one alternative embodiment, the above sample palmprint image generating method is also called an "identity-controllable palmprint generating method", and the overall flow of the method is as shown in FIG.

[0236] (1) Generate candidate palm print images 1110.

[0237] For example, based on observation of a large amount of real palm print data, the following rules for palm print texture have been realized: palm prints usually have 2 to 5 main lines, and for left hands, the main lines usually start in the upper left corner and end in the lower right corner, while for right hands, the reverse is true. The number of fine lines is even greater, and they are shallower and shorter than the main lines.

[0238] In some embodiments, a quadratic Bezier curve is used to parameterize the palm print to represent the palm print main lines and palm print thin lines, and three parameter points (one start point, one end point, and one adjustment point) are set in a 2D plane to complete the unique determination of one Bezier curve and simulate the palm print geometry.

[0239] Optionally, the above palm print texture parameterization determination process is used as a character control parameter to generate a Bezier character control curve.

[0240] In one alternative embodiment, the first and second regions corresponding to the direction of the palm print main line are determined based on a predetermined palm print main line distribution rule.

[0241] For illustrative purposes, a palmprint is taken as an example, and the distribution of the palmprint main lines in a palmprint has a certain regularity. As shown in Figure 3, which is a schematic diagram of the distribution of a palmprint, the palmprint main lines 320 are generally expressed with the upper left corner as the first main line positioning point and the lower right corner as the second main line positioning point, or the lower right corner is regarded as the first main line positioning point of the palmprint main lines 320 and the upper left corner as the second main line positioning point of the palmprint main lines 320. Optionally, the palmprint main lines are analyzed with the upper left corner as the first main line positioning point and the lower right corner as the second main line positioning point.

[0242] In one alternative embodiment, a first vertex and a second vertex that are diagonally related to each other are determined in a predetermined palmprint generation area, and a first area is determined within the palmprint generation area with the first vertex as the center and a first predetermined length as the radius.

[0243] For example, after determining the first vertex, the first region determined based on the first vertex is a sector region, where the center of the sector region is the first vertex and the radius of the sector region is a first predetermined length, or the sector region is regarded as a quarter circular region, where the center point of the circular region is the first vertex and the radius of the circular region is a first predetermined length.

[0244] Optionally, the first predetermined length may be a preset fixed value, or may be a numerical value determined based on the palm print generation area.

[0245] For example, if the palmprint generation area is a square area of ​​unit length, the circle center is the third vertex, and for the first main line positioning point, the coordinates of the first main line positioning point are defined as follows: TIFF2025528277000002.tif13170

[0246] Here, x indicates the horizontal axis coordinate and y indicates the vertical axis coordinate.

[0247] In one alternative embodiment, a second region is determined within the palm print generation region with the second vertex as its center and the second predetermined length as its radius.

[0248] Optionally, the process of determining the second region based on the second vertex is similar to the process of determining the first region based on the first vertex, where exemplarily, the second predetermined length may be the same as or different from the first predetermined length.

[0249] For example, as shown in FIG. 9, when the palmprint generation area is a square area of ​​unit length, the circle center is the third vertex, and for the first main line positioning point, the coordinates of the first main line positioning point are defined as follows: TIFF2025528277000003.tif14170

[0250] Here, x indicates the horizontal axis coordinate and y indicates the vertical axis coordinate.

[0251] Alternatively, if the second predetermined length is different from the first predetermined length, the second region formed with the second vertex as its center and the second predetermined length as its radius will have a different shape from the first region formed with the first vertex as its center and the first predetermined length as its radius. For example, if the palmprint generation region is a square region, and the second predetermined length is large and the first predetermined length is small, the second region corresponding to the second predetermined length will be larger than the first region corresponding to the first predetermined length.

[0252] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0253] In one alternative embodiment, after determining a first region, first data corresponding to the first main line positioning point is determined in the first region by random selection; after determining a second region, second data corresponding to the first main line positioning point is determined in the second region by random selection.

[0254] In some embodiments, the third region is determined according to the radian rule of the palm print main line and based on the positional relationship between the first main line positioning point and the second main line positioning point.

[0255] For example, the arc degree rule of the palm print main line indicates the curvature of the palm print main line, for example, the arc degree of the palm print main line is small, and the curvature of the palm print main line is smooth.

[0256] Optionally, a third region for generating main-line adjustment points is determined based on a positional relationship between the first main-line positioning point and the second main-line positioning point.

[0257] In one alternative embodiment, the first main line positioning point and the second main line positioning point are connected to obtain a target line segment.

[0258] For example, as shown in FIG. 12, in a palm print generation area 1210, a first main line positioning point 1220 (shown as an inverted triangle symbol) and a second main line positioning point 1230 (shown as a circle symbol) are determined, and the first main line positioning point 1220 and the second main line positioning point 1230 are connected to obtain a target line segment 1240.

[0259] Optionally, a rectangular area having a predetermined side length and centered on the midpoint of the target line segment is set as the third area.

[0260] For example, the predetermined side length may include a preset fixed value, or may include a value determined based on the palm print generation area.

[0261] For example, if the specified side length is a preset fixed value and the third area is determined based on the midpoint of a line segment, the third area is a rectangular area determined within the palmprint generation area with the midpoint of the line segment as the center and the preset fixed value as the side length. For example, the preset fixed values ​​include the length a of the rectangle and the width b of the rectangle, and the third area is obtained by setting the midpoint of the line segment as the center, the length a of the rectangle as the length of the third area, and the width b of the rectangle as the width of the third area.

[0262] Alternatively, when the specified side length is a value determined based on the palmprint generation area (for example, the side length is half the side length of the palmprint generation area, or half the side length of the target line segment of the palmprint generation area), and the first area is determined based on the midpoint of the line segment, the third area can be determined within the palmprint generation area with the midpoint of the line segment as the center and the value determined based on the palmprint generation area as the side length.

[0263] For example, as shown in FIG. 12, after obtaining a target line segment 1240, a third region 1250 is obtained by setting the midpoint of the target line segment 1240 as the center and setting a preset fixed value as the side length.

[0264] In one alternative embodiment, the coordinates of the center point of the target line segment connecting the first main line positioning point and the second main line positioning point are (x c ,y c), and the center point coordinates (x c ,y c ) as the center and the target line segment as the criterion, the process of determining the third region is as follows:

[0265] For example, if the length of the palmprint generation area is 1 unit length, the third area to be acquired is a square area, and the side length of the square area is set to 2 / 3 in advance, the third area will have a center point coordinate (x c ,y c ) and is a square area parallel to the target line segment with a side length of 2 / 3.

[0266] Optionally, the third region is enclosed by a method for determining a line equation, and the line equation is uniquely determined by the main line positioning point and the second main line positioning point. Illustratively, the line on which the target line segment connecting the first main line positioning point and the second main line positioning point is located is defined as line A, and line A is y=k1x+b1. Similarly, the center point coordinates (x c ,y c ) and define the line B perpendicular to the line A as y=k2x+b2.

[0267] Here, k1 indicates the slope corresponding to line A, b1 indicates the intercept corresponding to line A, k2 indicates the slope corresponding to line B, and b2 indicates the intercept corresponding to line B, and the relationship between k1, b1, k2, and b2 is as follows: TIFF2025528277000004.tif17170

[0268] That is, the line A and the center coordinate (x c ,y c ) the line B can be uniquely determined.

[0269] Optionally, as shown in FIG. 13, center point coordinates 1310 (x c ,y c), and based on the fact that the predetermined side length of the third region is 2 / 3, two lines that are parallel to line A and are 1 / 3 of the vertical distance from line A, namely line A11320 and line A21330, and two lines that are parallel to line B and are 1 / 3 of the vertical distance from line B, namely line B11340 and line B213350, are determined.

[0270] The equation of line A11320 is y=k1x+b3, the equation of line A21330 is y=k1x+b4, the equation of line B11340 is y=k2x+b5, and the equation of line B21350 is y=k2x+b6.

[0271] The above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0272] For example, as shown in Figure 13, based on the above four straight lines, line A11320, line A21330, line B11340, and line B21350, the values ​​of the four straight lines can be uniquely determined, a square area surrounded by the four straight lines can be determined, and this square area can be designated as the third area.

[0273] In one alternative embodiment, the adjustment point data is generated in a third region.

[0274] Optionally, after determining the third region, generate adjustment point data for adjusting the main line arc degree in the third region. Illustratively, the coordinate value range of the adjustment point data can be defined as follows: TIFF2025528277000005.tif30170

[0275] Here, k1 indicates the slope corresponding to the line A1 1320 and the line A2 1330, and b3 indicates the slope corresponding to the line A 13 b4 indicates the intercept corresponding to 120, b5 indicates the intercept corresponding to 1330 on line A2, and k6 indicates the intercept corresponding to 1330 on line B. 13 140 and the slope corresponding to line B21350, and b5 is the slope corresponding to line B 13b6 indicates the intercept corresponding to line B21350.

[0276] Optionally, a palm print main line is generated based on the first data, the second data and the adjustment point data.

[0277] Here, the palm print main line is a curve that sequentially connects the first main line positioning point, the main line adjustment point, and the second main line positioning point.

[0278] For example, the palmprint main line is a curve obtained by sequentially connecting the first main line positioning point, the main line adjustment point, and the second main line positioning point, where the coordinate positions of the first main line positioning point and the second main line positioning point are determined, and the main line adjustment point adjusts the arc degree of the line segment enclosed by the first main line positioning point and the second main line positioning point to obtain different palmprint main lines, that is, the position of the main line adjustment point and the formation of the palmprint main line are closely related.

[0279] In some embodiments, a candidate palm print image is generated that includes palm print main lines and palm print fine lines.

[0280] 14, a quadratic Bezier curve method is used to generate multiple candidate palm print images, which are also called Bezier curve images. For example, candidate palm print image 1410 includes three main palm print lines and five thin palm print lines, and candidate palm print image 1420 includes four main palm print lines and seven thin palm print lines.

[0281] (2) The candidate palmprint image 1110 and the palmprint image are input to a generative adversarial network 1120 for training, and a sample palmprint image is generated.

[0282] Illustratively, after obtaining the generated candidate palm print images and collected palm print images, the candidate palm print images and palm print images are input into a generative adversarial network as shown in FIG.

[0283] (1) A palmprint image is input to the encoder (f_encoder) 1510, and an image feature representation (feature_encoder) corresponding to the palmprint image is extracted. Here, the encoder is composed of a three-layer convolutional network to obtain an image feature representation corresponding to the palmprint image, and the feature dimension is 64.

[0284] (2) The extracted image feature representation is input as a conditional probability to a decoder (f_decoder) 1520 together with a candidate palmprint image (Bézier curve image bezier_img). The specific network structure of the decoder 1520 is as shown in FIG.

[0285] As shown in Figure 16, the decoder model structure is mainly composed of a three-layer downsampling network structure and a three-layer upsampling network structure.

[0286] The candidate palmprint image 1610 is first downsampled through three convolutional layers to extract a candidate feature representation (also called a Bezier feature representation) corresponding to the candidate palmprint image 1610, with a feature dimension of 64, the same as the image feature representation extracted in (1).

[0287] Optionally, a position-wise dot product operation is performed on the candidate feature representation and the image feature representation to obtain a fused feature representation 1620. Since the feature dimension of both the base image feature representation and the candidate feature representation is 64, the feature dimension of the fused feature representation 1620 is also 64.

[0288] The main purpose of this step is to allow the decoder to learn the muscle line information of the palm print image by constraining the candidate feature representation with the image feature representation corresponding to the collected palm print image. The fused feature representation 1620 is upsampled through a three-layer deconvolutional network to obtain a sample palm print image 1630 with muscles and palm print lines.

[0289] (3) The generated sample palmprint image (simulated palmprint image) and the palmprint image are mixed and input into the palmprint recognition model for training.

[0290] Illustratively, as shown in FIG. 15, after obtaining a sample palmprint image, the following formula is adopted to calculate the loss value L1_loss between the generated sample palmprint image and the palmprint image: TIFF2025528277000006.tif5170

[0291] Here, sample_img indicates the generated sample palmprint image, and gather_img indicates the gathered palmprint image.

[0292] For example, a sample palm print image is generated by the decoder 1520, and in the training stage of the palm print generation model, the decoder 1520 is trained using the above loss value L1_loss until the training condition is met.

[0293] (4) Using the feature extraction model 1530 obtained by training using the palmprint image, feature extraction is performed on the collected palmprint image and the generated sample palmprint image to obtain a palmprint feature representation feature_gt corresponding to the palmprint image and a sample feature representation feature_generated corresponding to the generated sample palmprint image. The cosine similarity sim between the palmprint feature representation and the sample feature representation is calculated, and (1-sim) is used as the identity constraint feature to constrain the identity consistency between the sample palmprint image and the palmprint image.

[0294] Illustratively, in the application stage of the palm print generation model, the generated candidate palm print image can be input to the decoder 1320 to obtain a generated sample palm print image. Optionally, the finally generated sample palm print image is realized as the image shown in Fig. 14, which includes not only palm print texture information but also muscle print information.

[0295] (3) Training the palmprint recognition model 1140 In one alternative embodiment, as shown in FIG. 17, a sample palm print image 1710 is used to perform model pre-training on the palm print recognition model.

[0296] Optionally, the model structure of the palmprint recognition model is realized as a mobile face detection network (mobilefacenet), and obtains a candidate recognition model 1720 based on model pre-training.

[0297] Then, a target palm print recognition model is obtained by fine-tuning the candidate recognition model 1720 using pre-collected palm print images 1730.

[0298] Please note that the above is an exemplary list, and the embodiments of the present application are not limited thereto.

[0299] As shown in the table below, this includes authentication data from the target palm print recognition model and authentication data obtained by other palm print recognition techniques.

[0300] [Table 1]

[0301] Table 1 shows a comparison of the authentication effectiveness in a public dataset between the palm print sample generation method according to the embodiment of the present application and conventional methods in the palm print recognition field, and the evaluation index is the recall rate (TPR: True Positive Rate).

[0302] [Table 2]

[0303] Table 2 shows a comparison of the authentication effectiveness in a service dataset between the palm print sample generation method according to the embodiment of the present application and the conventional method in the palm print recognition field, and the evaluation index is also TPR.

[0304] The baseline used here is ArcFace, a face recognition technology using a baseline method, and its backbone network is the mobile phone network (MobileFaceNet). As can be seen, the recognition results of the target palm print recognition model trained using the palm print sample generation method according to the embodiment of the present application are superior to those of ArcFace, both in the public dataset and the service dataset.

[0305] In summary, since the candidate palmprint images are obtained based on the palmprint main line distribution rules and palmprint fine line distribution rules, they have the characteristics of being large in quantity and diverse. At the same time, in order to make the candidate palmprint images closer to real palmprint images, constrained adjustment is performed on the generated candidate palmprint images using pre-collected palmprint images, which makes the palmprint distribution in the adjusted sample palmprint images more accurate, not only breaking through the limitation of a small number of palmprints but also improving the reliability of the sample palmprint images, thereby improving the robustness of the palmprint recognition model and helping to improve the palmprint recognition accuracy of the palmprint recognition model.

[0306] FIG. 18 is a block diagram of the structure of a sample palm print image generating device according to one exemplary embodiment of the present application. As shown in FIG. 18, the device includes: a main line generating module 1810 configured to generate at least one palm print main line based on a predetermined palm print main line distribution rule; a thin line generating module 1820 configured to determine a palm print distribution area in which the at least one palm print main line is located, and generate at least one palm print thin line within the palm print distribution area based on a predetermined palm print thin line distribution rule; an image acquisition module 1830 configured to obtain a candidate palm print image based on the at least one palm print main line and the at least one palm print thin line, wherein the candidate palm print image is labeled with a candidate image label, and the candidate image label is used to indicate a palm print distribution of the candidate palm print image, the palm print distribution including at least one of a palm print main line distribution and a palm print thin line distribution; an image generation module 1840 configured to input the candidate palmprint image into a pre-trained palmprint generation model to obtain a sample palmprint image, where the palmprint generation model is a model pre-trained using palmprint image labels of palmprint images collected in advance and the candidate image labels, where the palmprint image labels are used to indicate a palmprint distribution of the palmprint image, where the sample palmprint image is used to train a palmprint recognition model, and where the palmprint recognition model is used to perform palmprint recognition.

[0307] In one alternative embodiment, the palmprint generation model is a model obtained by performing model training on a candidate generation model.

[0308] As shown in FIG. 19, the device further comprises: The system further comprises a model training module 1850 configured to: acquire a palmprint image library, the palmprint image library having a plurality of collected palmprint images pre-stored therein, each of the plurality of palmprint images being labeled with a palmprint image label; extract features from the palmprint image to obtain an image feature representation corresponding to the palmprint image; extract features from the candidate palmprint image to obtain a candidate feature representation corresponding to the candidate palmprint image; fuse and upsample the candidate feature representation and the image feature representation using a candidate generative model to be trained to obtain a candidate sample palmprint image; train the candidate generative model using the candidate sample palmprint image to obtain the palmprint generation model, which is used to generate the sample palmprint image.

[0309] In one optional embodiment, the model training module 1850 is further configured to: determine a candidate palmprint image corresponding to the candidate sample palmprint image; use the candidate image label corresponding to the candidate palmprint image as a sample image label of the candidate sample palmprint image; obtain a loss value corresponding to the candidate sample palmprint image based on the distribution difference between the palmprint distribution indicated by the sample image label and the palmprint distribution indicated by the palmprint image label; and train the candidate generative model using the loss value to obtain the palmprint generative model.

[0310] In one alternative embodiment, the model training module 1850 is further configured to: perform a feature fusion operation on the candidate feature representation and the image feature representation using the candidate generative model to obtain a fused feature representation, which is used to represent a feature representation obtained by performing constraint adjustment on the candidate feature representation using at least one image feature representation; and perform an up-sampling operation on the fused feature representation to obtain the candidate sample palmprint image.

[0311] In an alternative embodiment, the apparatus further comprises: The method further comprises: obtaining a target feature extraction model, the target feature extraction model being a feature extraction model pre-trained using a plurality of palmprint images in the palmprint image library; and comprising: a label constraint module 1860 configured to extract features from the sample palmprint image and the at least one palmprint image using the target feature extraction model, obtain a first feature representation corresponding to the sample palmprint image and a second feature representation corresponding to the at least one palmprint image, respectively; and perform label constraint on the sample palmprint image based on feature differences between the first feature representation and the at least one second feature representation, wherein the label constraint is used to constrain identity consistency between the sample palmprint image and the at least one palmprint image.

[0312] In one optional embodiment, the label constraint module 1860 is further configured to perform a cosine similarity calculation on the first feature representation and the at least one second feature representation to obtain at least one feature similarity, and perform label constraint on the sample palmprint image based on the at least one feature similarity.

[0313] In an alternative embodiment, the apparatus further comprises: a palm print recognition model 1870 configured to train the palm print recognition model using the sample palm print image to obtain a candidate palm print recognition model, and to adjust the candidate palm print recognition model using at least one palm print image in the palm print image library to obtain a target palm print recognition model, wherein the target palm print recognition model is used to perform palm print recognition.

[0314] In one alternative embodiment, the image generation module 1840 is further configured to input the candidate palmprint image and the palmprint image into a pre-trained palmprint generation model, and perform constraint adjustment on the candidate palmprint image using the palmprint image to obtain the sample palmprint image.

[0315] In one alternative embodiment, the main line generation module 1810 is further configured to generate positioning point data according to a predetermined palm print main line distribution rule, where the positioning point data includes first data corresponding to first main line positioning points and second data corresponding to second main line positioning points; generate adjustment point data according to a palm print main line arc rule, where the main line adjustment points corresponding to the adjustment point data are used to control the arc degrees of the main line formed by the first main line positioning points and the second main line positioning points; and generate the at least one palm print main line according to the first data, the second data, and the adjustment point data, where the palm print main line is a curve connecting the first main line positioning points, the main line adjustment points, and the second main line positioning points in order.

[0316] In one optional embodiment, the thin line generation module 1820 is further configured to: determine a palm print distribution area in which the at least one palm print main line is located; determine positioning points of a plurality of palm print thin lines within the palm print distribution area based on a palm print thin line distribution rule; and generate the at least one palm print thin line within the palm print distribution area based on the positioning points of the plurality of palm print thin lines.

[0317] In summary, a pre-trained palmprint generation model is obtained using the palmprint image labels and candidate image labels of pre-collected palmprint images, and the palmprint generation model is used to generate sample palmprint images corresponding to the candidate palmprint images, thereby making the palmprint distribution in the sample palmprint images more accurate, not only breaking through the limitation of a small number of palmprints, but also improving the reliability of the sample palmprint images, thereby improving the robustness of the palmprint recognition model and helping to improve the palmprint recognition accuracy of the palmprint recognition model.

[0318] It should be noted that in the device for generating a sample palmprint image according to the above embodiment, the division of each of the above functional modules is merely an example, and in actual applications, the above functions can be completed by different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the above functions. Furthermore, the device for generating a sample palmprint image according to the above embodiment belongs to the same concept as the embodiment of the method for generating a sample palmprint image, and its specific implementation process can be referred to the embodiment of the method, and will not be repeated here.

[0319] 20 shows a schematic diagram of a server architecture according to one exemplary embodiment of the present application. The server 2000 includes a central processing unit (CPU) 2001, a system memory 2004 including a random access memory (RAM) 2002 and a read only memory (ROM) 2003, and a system bus 2005 connecting the system memory 2004 and the central processing unit 2001. The server 2000 further includes a non-volatile storage device 2006 for storing an operating system 2013, applications 2014, and other program modules 2015.

[0320] The non-volatile storage device 2006 is connected to the central processing unit 2001 via a non-volatile storage controller (not shown) connected to the system bus 2005. The non-volatile storage device 2006 and its associated computer-readable media provide non-volatile storage for the server 2000. That is, the non-volatile storage device 2006 may include a computer-readable medium (not shown) such as a hard disk or a CD-ROM (Compact Disc Read Only Memory) drive.

[0321] Without loss of generality, computer-readable media may include computer storage media and communication media. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other solid-state drive technology, CD-ROM, digital versatile disc (DVD) or other optical memory, magnetic tape cassettes, magnetic tape, magnetic disk storage, or other magnetic storage devices. Of course, those skilled in the art will understand that computer storage media is not limited to these several types. The system memory 2004 and nonvolatile storage device 2006 may be collectively referred to as memory.

[0322] According to various embodiments of the present application, the server 2000 may also be connected to and run on a remote computer on a network, such as the Internet, via a network, i.e., the server 2000 may be connected to a network 2012 via a network interface unit 2011 connected to the system bus 2005, or may use the network interface unit 2011 to connect to other types of networks or remote computer systems (not shown).

[0323] The memory further includes one or more programs stored in the memory and executed by the CPU.

[0324] An embodiment of the present application further provides a computing device, including a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the processor loads and executes the at least one instruction, at least one program, code set, or instruction set to implement the method for generating a sample palm print image according to each of the method embodiments described above.

[0325] An embodiment of the present application further provides a computer-readable storage medium storing at least one instruction, at least one program, code set or instruction set, which is loaded and executed by a processor to cause the processor to perform the method for generating a sample palm print image according to each of the method embodiments described above.

[0326] The present embodiment further provides a computer program product or a computer program, the computer program product including computer instructions stored in a computer-readable storage medium, wherein a processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the method for generating a sample palmprint image according to any one of the above embodiments of the present application.

[0327] For example, the computer-readable storage medium may include a read-only memory (ROM), a random access memory (RAM), a solid-state hard disk (SSD), an optical disk, etc. Here, the random access memory may include a resistive random access memory (ReRAM) and a dynamic random access memory (DRAM). The numbers of the above embodiments of the present application do not indicate the superiority or inferiority of the embodiments, but are provided for convenience of explanation.

[0328] Those skilled in the art can understand that all or part of the steps in the above embodiments can be implemented by hardware, and can also be realized by instructing related hardware by a program, and the program can be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk, an optical disk, etc.

[0329] The above are only optional examples of the present application, and do not limit the present application, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall all fall within the protection scope of the present application.

[0330] The various technical features in the above embodiments can be combined in any way, and for the sake of brevity, not all possible combinations of each technical feature in the above embodiments are described, but as long as there is no contradiction in the combinations between features, all should be considered within the scope described in this specification.

[0331] The above examples only represent some embodiments of the present application, and the description is specific and detailed, but should not be understood as a limitation on the scope of the claims of the present application. It should be noted that those skilled in the art can make some modifications and improvements without departing from the concept of the present application, and all of these should be considered to be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the attached claims.

Claims

1. 1. A computer device-implemented method for generating a sample palm print image, comprising: generating at least one palm print main line based on a predetermined palm print main line distribution rule; determining a palm print distribution area in which the at least one palm print main line is located, and generating at least one palm print thin line within the palm print distribution area based on a predetermined palm print thin line distribution rule; generating a candidate palm print image based on the at least one palm print main line and the at least one palm print thin line in the palm print distribution region, wherein the candidate palm print image is labeled with a candidate image label, the candidate image label is used to indicate a palm print distribution of the candidate palm print image, and the palm print distribution includes at least one of a palm print main line distribution and a palm print thin line distribution; and inputting the candidate palm print image into a pre-trained palm print generation model to obtain a sample palm print image, the palm print generation model being a model pre-trained using palm print image labels of pre-collected palm print images and the candidate image labels, the palm print image labels being used to indicate a palm print distribution of the pre-collected palm print images, and the sample palm print image being used to train a palm print recognition model.

2. the palmprint generation model is a model obtained by performing model training on a candidate generation model, The method for generating the sample palm print image includes the steps of: acquiring a palm print image library, wherein a plurality of collected palm print images are pre-stored in the palm print image library, and each of the plurality of palm print images is labeled with a palm print image label; extracting features from the palmprint image to obtain an image feature representation corresponding to the palmprint image; extracting features from the candidate palmprint image to obtain a candidate feature representation corresponding to the candidate palmprint image; fusing and upsampling the candidate feature representation and the image feature representation by a candidate generative model to be trained to obtain a candidate sample palmprint image; 2. The method for generating a sample palm print image according to claim 1, further comprising: training the candidate generative model using the candidate sample palm print image to obtain the palm print generative model, wherein the palm print generative model is used to generate the sample palm print image.

3. The step of training the candidate generative model using the candidate sample palm print images to obtain the palm print generative model includes: determining a candidate palm print image corresponding to the candidate sample palm print image, and using the candidate image label corresponding to the candidate palm print image as a sample image label for the candidate sample palm print image; obtaining a loss value corresponding to the candidate sample palmprint image based on a distribution difference between the palmprint distribution indicated by the sample image label and the palmprint distribution indicated by the palmprint image label; and training the candidate generative model using the loss value to obtain the palm print generative model.

4. The step of fusing the candidate feature representation and the image feature representation and performing upsampling using the candidate generative model to be trained to obtain a candidate sample palmprint image includes: performing a feature fusion process on the candidate feature representation and the image feature representation using the candidate generative model to obtain a fused feature representation, the fused feature representation being used to represent a feature representation obtained by performing constraint adjustment on the candidate feature representation using at least one image feature representation; and performing an upsampling process on the fused feature representation to obtain the candidate sample palm print image.

5. The method for generating the sample palm print image includes the steps of: obtaining a target feature extraction model, the target feature extraction model being a feature extraction model pre-trained using a plurality of palm print images in the palm print image library; extracting features from the sample palmprint image and the at least one palmprint image using the target feature extraction model to obtain a first feature representation corresponding to the sample palmprint image and a second feature representation corresponding to the at least one palmprint image, respectively; 5. The method of claim 1, further comprising: applying a label constraint to the sample palm print image based on feature differences between the first feature representation and at least one second feature representation, wherein the label constraint is used to constrain identity consistency between the sample palm print image and at least one palm print image.

6. performing label constraints on the sample palm print image based on feature differences between the first feature representation and at least one second feature representation, performing a cosine similarity calculation on the first feature representation and the at least one second feature representation to obtain at least one feature similarity; and performing label constraints on the sample palm print images based on the at least one feature similarity.

7. After inputting the candidate palm print image into a pre-trained palm print generation model to obtain a sample palm print image, the method for generating the sample palm print image includes: training the palm print recognition model using the sample palm print images to obtain candidate palm print recognition models; 7. The method for generating a sample palm print image according to claim 1, further comprising: training the candidate palm print recognition model using at least one palm print image in the palm print image library to obtain a target palm print recognition model, wherein the target palm print recognition model is used to perform palm print recognition.

8. The method for generating the sample palm print image includes the steps of:

8. The method for generating a sample palm print image according to claim 1, further comprising the steps of inputting the candidate palm print image and the palm print image into a pre-trained palm print generation model, and performing constraint adjustment on the candidate palm print image using the palm print image to obtain the sample palm print image.

9. The step of generating at least one palm print main line based on the predetermined palm print main line distribution rule includes: generating positioning point data based on a predetermined palm print main line distribution rule, the positioning point data including first data corresponding to a first main line positioning point and second data corresponding to a second main line positioning point; generating adjustment point data based on a rule of arc degree of a palm print main line, wherein the main line adjustment points corresponding to the adjustment point data are used to control the arc degree of a main line formed by the first main line positioning points and the second main line positioning points; 9. The method for generating a sample palm print image according to claim 1, further comprising: a step of generating the at least one palm print main line based on the first data, the second data, and the adjustment point data, wherein the palm print main line is a curve that sequentially connects the first main line positioning point, the main line adjustment point, and the second main line positioning point.

10. The step of generating at least one palm print thin line within the palm print distribution area based on the predetermined palm print thin line distribution rule includes: determining positioning points of a plurality of palm print thin lines within the palm print distribution area based on a predetermined palm print thin line distribution rule; and generating the at least one palm print thin line within the palm print distribution area based on the positioning points of the plurality of palm print thin lines.

11. A sample palm print image generating device, comprising: a main line generating module configured to generate at least one palm print main line based on a predetermined palm print main line distribution rule; a thin line generating module configured to determine a palm print distribution area in which the at least one palm print main line is located, and to generate at least one palm print thin line within the palm print distribution area based on a predetermined palm print thin line distribution rule; an image acquisition module configured to generate a candidate palm print image based on the at least one palm print main line and the at least one palm print thin line in the palm print distribution region, wherein the candidate palm print image is labeled with a candidate image label, the candidate image label is used to indicate a palm print distribution of the candidate palm print image, and the palm print distribution includes at least one of a palm print main line distribution and a palm print thin line distribution; an image generation module configured to input the candidate palmprint image into a pre-trained palmprint generation model to obtain a sample palmprint image, the palmprint generation model being a model pre-trained using palmprint image labels of pre-collected palmprint images and the candidate image labels, the palmprint image labels being used to indicate a palmprint distribution of the pre-collected palmprint images, and the sample palmprint image being used to train a palmprint recognition model.

12. a memory for storing at least one program; A processor that loads and executes the at least one program to implement the method for generating a sample palm print image according to any one of claims 1 to 10.

13. 11. A computer-readable storage medium storing at least one program that, when loaded and executed by a processor, causes the processor to perform the method for generating a sample palm print image according to any one of claims 1 to 10.

14. A computer program product comprising a computer program or instructions which, when executed by a processor, causes the processor to carry out the method for generating a sample palm print image according to any one of claims 1 to 10.

Images