Method and device for determining image recognition evaluation value, storage medium, and electronic device

Abstract

The present invention discloses a method and device for determining an image recognition evaluation value, a storage medium, and an electronic device, wherein the method comprises: obtaining a target quality parameter corresponding to a target image; determining a plurality of recognition reference values ​​corresponding to the target quality parameter based on a posterior probability model and a plurality of reference images; wherein the recognition reference value is used to indicate the accuracy of recognizing a target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object; determining a target recognition evaluation value of the target image based on the plurality of recognition reference values, wherein the target recognition evaluation value is used to indicate the average accuracy of recognizing the target object in the target image. The above scheme is adopted to solve the problems in the related art that the subjectivity of artificially annotating the recognition evaluation value of the target image is too strong, and the recognition evaluation value often cannot accurately measure the quality of the target image.

Description

Technical Field

[0001] The present invention relates to the field of communications, and in particular to a method and device for determining an image recognition evaluation value, a storage medium, and an electronic device. Background Art

[0002] Person re-identification technology has always been a hot topic in computer vision research and application. The prerequisite for effectively matching pedestrian targets in videos with pedestrian targets in the database is that the contours of each part of the target image in the input model are clear and complete, the target posture is good, and the target image quality is good. Therefore, judging the quality level of the target image and screening out targets with complete and clearly visible key parts can effectively reduce the demand for computing resources and improve the accuracy and efficiency of person re-identification.

[0003] The main method for judging the optimal level of human images is to collect a large amount of labeled data to train the relevant classification network model and output the optimal score of the target. However, in actual scenes, the environment is often complex, the postures of the targets are diverse, and the optimal scores of human annotations are often easily affected by the subjective factors of the annotators themselves, which cannot accurately reflect the impact of the overall quality of the target on the accuracy of the pedestrian re-identification task. Figure 3 As shown in the figure, the target completeness is a dimension of the overall quality of the target image, and the occlusion degree of the upper body of the human body in the figure will continue to change in multiple consecutive frames of images. The preferred score corresponding to this subtle change is often difficult to give through manually defined standards, and it is also difficult to describe the impact of the occlusion degree on the final pedestrian re-identification even if it is given.

[0004] In view of the problems in related technologies such as the subjectivity of manually annotating recognition evaluation values ​​of target images and the inability of recognition evaluation values ​​to accurately measure the quality of target images, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for determining an image recognition evaluation value, a storage medium, and an electronic device, so as to at least solve the problems in the related art that the recognition evaluation value manually annotated on the target image is too subjective, and the recognition evaluation value often cannot accurately measure the quality of the target image.

[0006] According to one embodiment of the present invention, a method for determining an image recognition evaluation value is provided, comprising: obtaining a target quality parameter corresponding to a target image; determining a plurality of recognition reference values ​​corresponding to the target quality parameter based on a posterior probability model and a plurality of reference images; wherein the recognition reference value is used to indicate the accuracy of recognizing a target object in the target image based on a corresponding reference image and the target quality parameter, the reference image and the target image containing the same target object; determining a target recognition evaluation value of the target image based on the plurality of recognition reference values, wherein the target recognition evaluation value is used to indicate the average accuracy of recognizing the target object in the target image.

[0007] In an exemplary embodiment, the target quality parameter includes at least one of the following information: an image quality level of the target image, a posture level of a target object contained in the target image, and a completeness of the target object.

[0008] In an exemplary embodiment, obtaining a target quality parameter corresponding to a target image includes: obtaining a plurality of reference images in a database, wherein the reference image and the target image contain the same target object; combining the target image and the plurality of reference images into a plurality of sets, wherein the quality parameters of the plurality of sets are the same, and the sets include: the target image and a portion of the reference images; obtaining the quality parameter of any one of the plurality of sets, and using the quality parameter of any one of the sets as the target quality parameter of the target image.

[0009] In an exemplary embodiment, determining multiple identification reference values ​​corresponding to the quality parameters according to a posterior probability model includes: inputting the multiple sets into the posterior probability model respectively according to a preset order; instructing the posterior probability model to determine the identification reference value corresponding to each set in the multiple sets according to the quality parameters of the multiple sets, so as to obtain multiple identification reference values, and using the multiple identification reference values ​​as the multiple identification reference values ​​corresponding to the target quality parameters.

[0010] In an exemplary embodiment, before determining multiple identification reference values ​​corresponding to the target quality parameter based on the posterior probability model and multiple reference images, the method also includes: constructing a first set based on a first reference image and a first test image corresponding to the first reference image, and determining a first quality parameter of the first set, wherein the first target object of the first reference image is the same as the second target object of the second test image; determining a matching degree between the first target object in the first reference image and the second target object in the first test image through a pedestrian re-identification algorithm; and determining a first identification reference value of the first quality parameter based on the matching degree.

[0011] In an exemplary embodiment, after determining the first identification reference value of the first quality parameter according to the matching degree, the method further includes: constructing a Gaussian mixture model by the following formula: Where 0≤π k ≤1, μ k It is used to indicate the mean of the kth component of the Gaussian mixture model, ∑ k Used to indicate the mean and variance of the kth component of the Gaussian mixture model, q is the first quality parameter, r is the first identification reference value, wherein the Gaussian mixture model is used to indicate the probability distribution of the first quality parameter and the first identification reference value.

[0012] In an exemplary embodiment, after constructing the Gaussian mixture model by the following formula, the method further includes: solving the Gaussian mixture model by a preset algorithm; determining the posterior probability model of the identification reference value according to the solved Gaussian mixture model, wherein the posterior probability model is:

[0013] According to another embodiment of the present invention, there is also provided a device for determining an image recognition evaluation value, comprising: an acquisition module, for acquiring a target quality parameter corresponding to a target image; a first determination module, for determining, based on a posterior probability model and based on a plurality of reference images, a plurality of recognition reference values ​​corresponding to the target quality parameter; wherein the recognition reference value is used to indicate the accuracy of recognizing a target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object; and a second determination module, for determining a target recognition evaluation value of the target image based on the plurality of recognition reference values, wherein the target recognition evaluation value is used to indicate the average accuracy of recognizing the target object in the target image.

[0014] In an exemplary embodiment, the acquisition module is also used to acquire multiple reference images in a database, wherein the reference image and the target image contain the same target object; combine the target image and the multiple reference images into multiple sets, wherein the quality parameters of the multiple sets are the same, and the sets include: the target image and part of the reference images; acquire the quality parameters of any one of the multiple sets, and use the quality parameters of any one of the sets as the target quality parameters of the target image.

[0015] According to yet another embodiment of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the steps of any one of the above method embodiments when run.

[0016] According to yet another embodiment of the present invention, there is provided an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0017] According to the present invention, a target quality parameter corresponding to a target image is obtained; a plurality of recognition reference values ​​corresponding to the target quality parameter are determined according to a posterior probability model and a plurality of reference images; wherein the recognition reference value is used to indicate the accuracy of recognizing a target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object; a target recognition evaluation value of the target image is determined according to the plurality of recognition reference values, wherein the target recognition evaluation value is used to indicate the average accuracy of recognizing the target object in the target image, that is, the recognition reference value corresponding to the target quality parameter of the target image is determined according to the posterior probability model, and the target recognition evaluation value of the target image is determined according to the plurality of recognition reference values. The above scheme is adopted to solve the problems in the related art that the subjectivity of artificially annotating the recognition evaluation value of the target image is too strong, and the recognition evaluation value often cannot accurately measure the quality of the target image. The present invention obtains the recognition evaluation value of the target image by fusing the dimensions of the image quality, the target posture and the target integrity through the posterior probability model, and can better reflect the degree of adaptation of the target image to the downstream model of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] Figure 1 It is a hardware structure block diagram of a computer terminal of a method for determining an image recognition evaluation value according to an embodiment of the present invention;

[0020] Figure 2 is a flowchart of a method for determining an image recognition evaluation value according to an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of a target image according to an optional embodiment of the present invention;

[0022] Figure 4 A schematic diagram of a method for determining an image recognition evaluation value according to an optional embodiment of the present invention;

[0023] Figure 5 A schematic diagram of quality parameters of a method for determining an image recognition evaluation value according to an optional embodiment of the present invention;

[0024] Figure 6 4 is a structural block diagram of a device for determining an image recognition evaluation value according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0026] It should be noted that the terms "historical", "second" and the like in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0027] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal or a similar computing device. Taking running on a computer terminal as an example, Figure 1 FIG. 1 is a hardware structure block diagram of a computer terminal for determining an image recognition evaluation value according to an embodiment of the present invention. Figure 1 As shown, the computer terminal may include one or more ( Figure 1 Only one is shown in the figure) processor 102 (processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. In an exemplary embodiment, the computer terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Equivalent functions or comparisons shown Figure 1 A different configuration with more features is shown.

[0028] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the method for determining the image recognition evaluation value of the computer terminal in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0029] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0030] In this embodiment, a method for determining an image recognition evaluation value is provided, which is applied to the above-mentioned computer terminal. Figure 2 : is a flowchart of a method for determining an image recognition evaluation value according to an embodiment of the present invention, the process comprising the following steps:

[0031] Step S202, obtaining a target quality parameter corresponding to the target image;

[0032] Step S204, determining a plurality of identification reference values ​​corresponding to the target quality parameter according to the posterior probability model and a plurality of reference images; wherein the identification reference value is used to indicate the accuracy of identifying the target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object;

[0033] Step S206: determining a target recognition evaluation value of the target image according to the multiple recognition reference values, wherein the target recognition evaluation value is used to indicate an average accuracy level of recognizing the target object in the target image.

[0034] Through the above steps, a target quality parameter corresponding to the target image is obtained; a plurality of recognition reference values ​​corresponding to the target quality parameter are determined according to the posterior probability model and a plurality of reference images; wherein the recognition reference value is used to indicate the accuracy of recognizing the target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object; a target recognition evaluation value of the target image is determined according to the plurality of recognition reference values, wherein the target recognition evaluation value is used to indicate the average accuracy of recognizing the target object in the target image, that is, the recognition reference value corresponding to the target quality parameter of the target image is determined according to the posterior probability model, and the target recognition evaluation value of the target image is determined according to the plurality of recognition reference values. The above scheme is adopted to solve the problems in the related art that the subjectivity of artificially annotating the recognition evaluation value of the target image is too strong, and the recognition evaluation value often cannot accurately measure the quality of the target image. The present invention obtains the recognition evaluation value of the target image by fusing the dimensions of the image quality, the target posture and the target integrity through the posterior probability model, and can better reflect the degree of adaptation of the target image to the downstream model of the algorithm.

[0035] It should be noted that the labels of the target objects in the reference image and the target image are consistent, that is, the target objects in the reference image and the target image are the same target objects, but other information such as posture and size may be inconsistent.

[0036] In an exemplary embodiment, the target quality parameter includes at least one of the following information: an image quality level of the target image, a posture level of a target object contained in the target image, and a completeness of the target object.

[0037] That is to say, the target quality parameters in the embodiment of the present invention are determined based on one or more of the image quality level of the target image, the posture level of the target object contained in the target image, and the degree of integrity of the target object. It should be noted that in order to more objectively reflect the target quality parameters of the target image, the target quality parameters are determined based on the image quality level of the target image, the posture level of the target object contained in the target image, and the degree of integrity of the target object.

[0038] In an exemplary embodiment, obtaining a target quality parameter corresponding to a target image includes: obtaining a plurality of reference images in a database, wherein the reference image and the target image contain the same target object; combining the target image and the plurality of reference images into a plurality of sets, wherein the quality parameters of the plurality of sets are the same, and the sets include: the target image and a portion of the reference images; obtaining the quality parameter of any one of the plurality of sets, and using the quality parameter of any one of the sets as the target quality parameter of the target image.

[0039] That is, the target quality parameters corresponding to the target image are obtained in the following manner: determine the target object in the target image, determine a first target object that is identical to the target object in the reference images in the database, determine multiple reference images corresponding to the first target object, randomly select a first preset number of reference images and their corresponding target images from the multiple reference images as a set, determine a second preset number of multiple sets from the multiple reference images, the image quality parameters of the multiple sets are exactly the same, and the image quality parameters of the sets are used as the target quality parameters corresponding to the target image.

[0040] In an exemplary embodiment, determining multiple identification reference values ​​corresponding to the quality parameters according to a posterior probability model includes: inputting the multiple sets into the posterior probability model respectively according to a preset order; instructing the posterior probability model to determine the identification reference value corresponding to each set in the multiple sets according to the quality parameters of the multiple sets, so as to obtain multiple identification reference values, and using the multiple identification reference values ​​as the multiple identification reference values ​​corresponding to the target quality parameters.

[0041] Specifically, the quality parameter q corresponding to the target image is obtained, and the reference value r is identified by the following solution:

[0042] r * = arg max r P(q, r) obtains multiple identification reference values ​​r, for r = [r1, r2, ..., r n ]∈R n The target recognition evaluation value s of the target image can be obtained by taking the average value of all elements in the target recognition evaluation value s. The target recognition evaluation value s is used to indicate the average accuracy rate of identifying the target object in the target image.

[0043] In an exemplary embodiment, before determining multiple identification reference values ​​corresponding to the target quality parameter based on the posterior probability model and multiple reference images, the method also includes: constructing a first set based on a first reference image and a first test image corresponding to the first reference image, and determining a first quality parameter of the first set, wherein the first target object of the first reference image is the same as the second target object of the second test image; determining a matching degree between the first target object in the first reference image and the second target object in the first test image through a pedestrian re-identification algorithm; and determining a first identification reference value of the first quality parameter based on the matching degree.

[0044] Before determining the multiple identification reference values ​​corresponding to the quality parameter according to the posterior probability model, it is also necessary to determine the posterior probability model, determine the posterior probability model by the first identification reference value of the first quality parameter, and specifically determine the first identification reference value of the first quality parameter in the following manner: Assuming that there are N pieces of pedestrian re-identification data in total, the pedestrian re-identification data includes: a reference image and a test image, a reference image with a preset value determined from the reference image and a test image with a preset value corresponding to the reference object, wherein the first reference image is the same as the target object in the first test image, the second reference image is the same as the target object in the second test image, and so on, the determined first reference image and the first test image, the second reference image and the second test image, ..., are combined into a set, and the first quality parameter of the set is determined according to the reference image and the test image in the set, and the matching degree of the reference image and the test image in the set is determined by the preset pedestrian re-identification algorithm to determine the first identification reference value corresponding to the first quality parameter. For N pieces of pedestrian re-identification data, multiple sets are determined according to the N pieces of pedestrian re-identification data, and the first quality parameters of the multiple sets are completely the same, and then the first identification reference values ​​of the first quality parameters of the multiple sets are obtained.

[0045] Furthermore, multiple sets are determined based on N pieces of pedestrian re-identification data, and the second quality parameters of the multiple sets are exactly the same, thereby obtaining second identification reference values ​​of the second quality parameters of the multiple sets; multiple sets are determined based on N pieces of pedestrian re-identification data, and the third quality parameters of the multiple sets are exactly the same, thereby obtaining third identification reference values ​​of the third quality parameters of the multiple sets…, determining a preset number of quality parameters and the identification reference values ​​corresponding to the quality parameters.

[0046] In an exemplary embodiment, after determining the first identification reference value of the first quality parameter according to the matching degree, a Gaussian mixture model is constructed by the following formula: Where 0≤π k ≤1, μ k It is used to indicate the mean of the kth component of the Gaussian mixture model, ∑ k Used to indicate the mean and variance of the kth component of the Gaussian mixture model, q is the first quality parameter, r is the first identification reference value, wherein the Gaussian mixture model is used to indicate the probability distribution of the first quality parameter and the first identification reference value.

[0047] The obtained quality parameters and the identification reference values ​​corresponding to the quality parameters are used to construct a Gaussian mixture model to describe the probability distribution of the quality parameters and the identification reference values ​​corresponding to the quality parameters. The formula is as follows:

[0048]

[0049] Where 0≤π k ≤1 and μ k ∈R 2m+n is the mean of the kth component of the Gaussian mixture model, ∑ k is the kth variance of the Gaussian mixture model.

[0050] In an exemplary embodiment, after constructing the Gaussian mixture model by the following formula, the method further includes: solving the Gaussian mixture model by a preset algorithm; determining the posterior probability model of the identification reference value according to the solved Gaussian mixture model, wherein the posterior probability model is:

[0051] Use the EM algorithm to solve the Gaussian mixture model and obtain the posterior probability model for identifying the reference value:

[0052] In the posterior probability model, when P(r|q) reaches the maximum value, the corresponding identification reference value is the identification reference value that can be obtained under the quality parameter, and its solution is recorded as: * = arg max r P(q,r).

[0053] In order to better understand the process of the method for determining the image recognition evaluation value, the process of the method for determining the image recognition evaluation value is described below in conjunction with an optional embodiment, but is not intended to limit the technical solution of the embodiment of the present invention.

[0054] In the related art, the patent with publication number CN111696128A discloses a multi-task deep learning system including target detection, target tracking and image optimization to solve the problem of image optimization and storage. In this method, several frames of video sequence are first input, the target to be tracked in the current frame is detected by the SSD detector, the target to be tracked is tracked by the multi-target tracking algorithm based on the Hungarian algorithm, and data association, assignment and labeling are performed according to the target to be tracked marked in the previous frame. Finally, the candidate key target images of the current frame are scored and the preferred target image is updated. The image optimization part uses a twin network to obtain the quality score of the target image, and selects the image with a high score as the preferred image.

[0055] However, the above scheme only considers the image quality of the image, and lacks consideration of other influencing factors such as human posture and human integrity. In addition, the target recognition evaluation value of the target image is too subjective, and the target recognition evaluation value often cannot accurately measure the quality of the target image.

[0056] In the related art, the patent with publication number CN111241927A discloses a cascaded face image optimization method, including: face detection and coordinate estimation through multi-task cascade convolutional neural network; filtering pictures with resolution lower than the threshold; filtering pictures with crooked face posture; calculating the comprehensive score of the picture through resolution score, face head posture score, face occlusion score, expression score, clarity score, and light intensity score, and the picture with the highest comprehensive score is the optimal face; tracking the face image, pushing the optimal face, when the optimal face is updated or the optimal face is not updated within a preset number of frames of the same face and the comprehensive score is greater than the preset score, push the optimal face to the recognition module. Among them, the occlusion score uses the deep learning network to directly output an occlusion score of 0 to 10 to determine the degree of face occlusion.

[0057] However, in the occlusion scoring stage of the above scheme, the neural network often directly outputs the integrity score of the target, and the construction of the integrity score label is often manually calibrated according to a set of standards. The score given according to this set of standards often cannot accurately measure the image.

[0058] Therefore, the embodiments of the present invention provide a method and device for determining an image recognition evaluation value, a storage medium, and an electronic device, so as to at least solve the problems in the related art that the recognition evaluation value manually annotated on the target image is too subjective, and the recognition evaluation value often cannot accurately measure the quality of the target image.

[0059] Figure 4 is a schematic diagram of a method for determining an image recognition evaluation value according to an optional embodiment of the present invention, such as Figure 4 As shown, the specific steps for determining the image recognition evaluation value are as follows:

[0060] Step S401: performing five-level labeling of image quality, human posture and human integrity on pedestrian re-identification data;

[0061] Specifically, the pedestrian re-identification data includes probe data (equivalent to the test image in the above embodiment) and gallery data (equivalent to the reference image in the above embodiment). Multiple target images in the pedestrian re-identification data are labeled with five levels of image quality (equivalent to the image quality in the above embodiment), human posture (equivalent to the posture of the target object in the above embodiment) and human integrity (equivalent to the integrity of the target object in the above embodiment). The five levels of the three are good, better, medium, poor and poor, which are mapped to five scores of 5, 4, 3, 2 and 1 respectively. The specific example is as follows: Figure 5 As shown, for all pedestrian re-identification data, the image quality parameters corresponding to the corresponding probe images can be obtained Image quality parameters corresponding to gallery images Among them, a, b and c correspond to the grade scores of the three dimensions of image quality, human posture and human integrity respectively.

[0062] Step S402: constructing a relationship between the image quality parameter and the recognition performance (equivalent to the recognition reference value in the above embodiment);

[0063] Assuming that there are N total person re-identification data, randomly select m first gallery images and their corresponding m first probe images as the first image set, where the first gallery image and the first probe image are used to indicate the same target object, then the image quality parameter of the first image set can be obtained: Through the preset person re-identification algorithm, it is determined whether the first gallery image and the first probe image in the first image set match each other, in other words, the corresponding recognition performance r is obtained. For N pieces of person re-identification data, multiple image sets consisting of a sufficient number of m pairs of probe-gallery images can be obtained. The image quality parameters q of the multiple image sets are exactly the same. Selecting n small sets can obtain the following table:

[0064]

[0065] As shown in the table above, we can obtain n small sets, each of which contains m probe-gallery image pairs, and the quality parameter q is the same, but the probe-gallery image pairs in multiple sets are different). The corresponding recognition performance of each small set is r i , so r(j)=[r1,r2,…,r n ]∈R n , simply denoted as r, represents the recognition performance of the person re-identification algorithm for a specific quality parameter q.

[0066] Different quality parameters q correspond to different recognition performances r. The relationship between the quality parameter q and the recognition performance is represented by a probability density function P(q, r).

[0067] Step S403: constructing a mixed Gaussian model and solving it;

[0068] Assuming that we can obtain enough different (q, r) pairs, we can construct a Gaussian mixture model to describe the overall probability distribution of (q, r), as follows:

[0069] Where 0≤π k ≤1 and μ k ∈R 2m+n is the mean of the kth component of the Gaussian mixture model, ∑k is the variance of the kth component of the Gaussian mixture model. The Gaussian mixture model is solved by the EM algorithm. The posterior probability model of the recognition performance r is obtained:

[0070]

[0071] In this posterior probability model, when P(r|q) reaches the maximum value, the corresponding r is the pedestrian re-identification performance that can be obtained under the quality parameter q, and its solution is recorded as: * = arg max r P(q,r).

[0072] Step S404: inputting the quality parameter corresponding to the image to be predicted to obtain the preferred score of the image (equivalent to the identification reference value in the above embodiment).

[0073] Input the quality parameter q corresponding to the image to be predicted and solve r * = arg max r P(q, r) obtains the recognition performance r, for r = [r1, r2, ..., r n ]∈R n The optimal score of the image is obtained by taking the average value of all elements in , which is also the average recognition performance of the image in the pedestrian re-identification algorithm.

[0074] According to the present invention, a target quality parameter corresponding to a target image is obtained; a plurality of recognition reference values ​​corresponding to the target quality parameter are determined according to a posterior probability model and a plurality of reference images; wherein the recognition reference value is used to indicate the accuracy of recognizing a target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object; a target recognition evaluation value of the target image is determined according to the plurality of recognition reference values, wherein the target recognition evaluation value is used to indicate the average accuracy of recognizing the target object in the target image, that is, the recognition reference value corresponding to the target quality parameter of the target image is determined according to the posterior probability model, and the target recognition evaluation value of the target image is determined according to the plurality of recognition reference values. The above scheme is adopted to solve the problems in the related art that the subjectivity of artificially annotating the recognition evaluation value of the target image is too strong, and the recognition evaluation value often cannot accurately measure the quality of the target image. The present invention obtains the recognition evaluation value of the target image by fusing the dimensions of the image quality, the target posture and the target integrity through the posterior probability model, and can better reflect the degree of adaptation of the target image to the downstream model of the algorithm.

[0075] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0076] In this embodiment, a device for determining an image recognition evaluation value is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be repeated hereafter. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0077] Figure 6 is a structural block diagram of a device for determining an image recognition evaluation value according to an embodiment of the present invention; Figure 6 As shown, including:

[0078] An acquisition module 62, used to acquire a target quality parameter corresponding to a target image;

[0079] A first determination module 64 is used to determine a plurality of identification reference values ​​corresponding to the target quality parameter based on a plurality of reference images according to a posterior probability model; wherein the identification reference value is used to indicate the accuracy of identifying the target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object;

[0080] The second determination module 66 is used to determine a target recognition evaluation value of the target image according to the multiple recognition reference values, wherein the target recognition evaluation value is used to indicate an average accuracy level of recognizing the target object in the target image.

[0081] Through the above-mentioned device, a target quality parameter corresponding to the target image is obtained; a plurality of recognition reference values ​​corresponding to the target quality parameter are determined according to the posterior probability model and a plurality of reference images; wherein the recognition reference value is used to indicate the accuracy of recognizing the target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object; a target recognition evaluation value of the target image is determined according to the plurality of recognition reference values, wherein the target recognition evaluation value is used to indicate the average accuracy of recognizing the target object in the target image, that is, the recognition reference value corresponding to the target quality parameter of the target image is determined according to the posterior probability model, and the target recognition evaluation value of the target image is determined according to the plurality of recognition reference values. The above-mentioned scheme solves the problems in the related art that the subjectivity of artificially annotating the recognition evaluation value of the target image is too strong, and the recognition evaluation value often cannot accurately measure the quality of the target image. The present invention obtains the recognition evaluation value of the target image by fusing the dimensions of the image quality, the target posture and the target integrity through the posterior probability model, and can better reflect the degree of adaptation of the target image to the downstream model of the algorithm.

[0082] It should be noted that the labels of the target objects in the reference image and the target image are consistent, that is, the target objects in the reference image and the target image are the same target objects, but other information such as posture and size may be inconsistent.

[0083] In an exemplary embodiment, the target quality parameter includes at least one of the following information: an image quality level of the target image, a posture level of a target object contained in the target image, and a completeness of the target object.

[0084] That is to say, the target quality parameters in the embodiment of the present invention are determined based on one or more of the image quality level of the target image, the posture level of the target object contained in the target image, and the degree of integrity of the target object. It should be noted that in order to more objectively reflect the target quality parameters of the target image, the target quality parameters are determined based on the image quality level of the target image, the posture level of the target object contained in the target image, and the degree of integrity of the target object.

[0085] In an exemplary embodiment, the acquisition module is also used to acquire multiple reference images in a database, wherein the reference image and the target image contain the same target object; combine the target image and the multiple reference images into multiple sets, wherein the quality parameters of the multiple sets are the same, and the sets include: the target image and part of the reference images; acquire the quality parameters of any one of the multiple sets, and use the quality parameters of any one of the sets as the target quality parameters of the target image.

[0086] That is, the acquisition module obtains the target quality parameters corresponding to the target image in the following manner: determining the target object in the target image, determining a first target object that is identical to the target object in the reference images in the database, determining multiple reference images corresponding to the first target object, optionally selecting a first preset number of reference images and their corresponding target images from the multiple reference images as a set, determining a second preset number of multiple sets from the multiple reference images, the image quality parameters of the multiple sets are exactly the same, and the image quality parameters of the sets are used as the target quality parameters corresponding to the target image.

[0087] In an exemplary embodiment, the first determination module is also used to input the multiple sets into the posterior probability model respectively in a preset order; instruct the posterior probability model to determine the identification reference value corresponding to each set in the multiple sets according to the quality parameters of the multiple sets, so as to obtain multiple identification reference values, and use the multiple identification reference values ​​as the multiple identification reference values ​​corresponding to the target quality parameters.

[0088] Specifically, the quality parameter q corresponding to the target image is obtained, and the reference value r is identified by the following solution:

[0089] r * = arg max r P(q, r) obtains multiple identification reference values ​​r, for r = [r1, r2, ..., r n ]∈R n The target recognition evaluation value s of the target image can be obtained by taking the average value of all elements in the target recognition evaluation value s. The target recognition evaluation value s is used to indicate the average accuracy rate of identifying the target object in the target image.

[0090] In an exemplary embodiment, the first determination module is also used to construct a first set based on a first reference image and a first test image corresponding to the first reference image, and determine a first quality parameter of the first set, wherein the first target object of the first reference image is the same as the second target object of the second test image; determine the matching degree between the first target object in the first reference image and the second target object in the first test image through a pedestrian re-identification algorithm; and determine a first recognition reference value of the first quality parameter based on the matching degree.

[0091] Before determining the multiple identification reference values ​​corresponding to the quality parameter according to the posterior probability model, it is also necessary to determine the posterior probability model, determine the posterior probability model by the first identification reference value of the first quality parameter, and specifically determine the first identification reference value of the first quality parameter in the following manner: Assuming that there are N pieces of pedestrian re-identification data in total, the pedestrian re-identification data includes: a reference image and a test image, a reference image with a preset value determined from the reference image and a test image with a preset value corresponding to the reference object, wherein the first reference image is the same as the target object in the first test image, the second reference image is the same as the target object in the second test image, and so on, the determined first reference image and the first test image, the second reference image and the second test image, ..., are combined into a set, and the first quality parameter of the set is determined according to the reference image and the test image in the set, and the matching degree of the reference image and the test image in the set is determined by the preset pedestrian re-identification algorithm to determine the first identification reference value corresponding to the first quality parameter. For N pieces of pedestrian re-identification data, multiple sets are determined according to the N pieces of pedestrian re-identification data, and the first quality parameters of the multiple sets are completely the same, and then the first identification reference values ​​of the first quality parameters of the multiple sets are obtained.

[0092] Furthermore, multiple sets are determined based on N pieces of pedestrian re-identification data, and the second quality parameters of the multiple sets are exactly the same, thereby obtaining second identification reference values ​​of the second quality parameters of the multiple sets; multiple sets are determined based on N pieces of pedestrian re-identification data, and the third quality parameters of the multiple sets are exactly the same, thereby obtaining third identification reference values ​​of the third quality parameters of the multiple sets…, determining a preset number of quality parameters and the identification reference values ​​corresponding to the quality parameters.

[0093] In an exemplary embodiment, the first determination module is further configured to construct a Gaussian mixture model using the following formula: Where 0≤π k ≤1, μ k It is used to indicate the mean of the kth component of the Gaussian mixture model, ∑ k Used to indicate the mean and variance of the kth component of the Gaussian mixture model, q is the first quality parameter, r is the first identification reference value, wherein the Gaussian mixture model is used to indicate the probability distribution of the first quality parameter and the first identification reference value.

[0094] The obtained quality parameters and the identification reference values ​​corresponding to the quality parameters are used to construct a Gaussian mixture model to describe the probability distribution of the quality parameters and the identification reference values ​​corresponding to the quality parameters. The formula is as follows:

[0095]

[0096] Where 0≤π k≤1 and μ k ∈R 2m+n is the mean of the kth component of the Gaussian mixture model, ∑ k is the kth variance of the Gaussian mixture model.

[0097] In an exemplary embodiment, the first determination module is further used to solve the Gaussian mixture model by a preset algorithm; determine the posterior probability model of the identification reference value according to the solved Gaussian mixture model, wherein the posterior probability model is:

[0098] Specifically, the EM algorithm is used to solve the Gaussian mixture model to obtain the posterior probability model for identifying the reference value:

[0099] In the posterior probability model, when P(r|q) reaches the maximum value, the corresponding identification reference value is the identification reference value that can be obtained under the quality parameter, and its solution is recorded as: * = arg max r P(q,r).

[0100] In order to better understand the process of the method for determining the image recognition evaluation value, the process of the method for determining the image recognition evaluation value is described below in conjunction with an optional embodiment, but is not intended to limit the technical solution of the embodiment of the present invention.

[0101] An embodiment of the present invention further provides a storage medium, which includes a stored program, wherein the program executes any of the above methods when it is run.

[0102] In an exemplary embodiment, in this embodiment, the above storage medium may be configured to store a computer program for performing the following steps:

[0103] S1, obtaining target quality parameters corresponding to the target image;

[0104] S2, determining a plurality of identification reference values ​​corresponding to the target quality parameter according to a posterior probability model and a plurality of reference images; wherein the identification reference value is used to indicate the accuracy of identifying the target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object;

[0105] S3, determining a target recognition evaluation value of the target image according to the multiple recognition reference values, wherein the target recognition evaluation value is used to indicate an average accuracy level of recognizing the target object in the target image.

[0106] In an exemplary embodiment, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store computer programs.

[0107] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0108] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0109] In an exemplary embodiment, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0110] S1, obtaining target quality parameters corresponding to the target image;

[0111] S2, determining a plurality of identification reference values ​​corresponding to the target quality parameter according to a posterior probability model and a plurality of reference images; wherein the identification reference value is used to indicate the accuracy of identifying the target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object;

[0112] S3, determining a target recognition evaluation value of the target image according to the multiple recognition reference values, wherein the target recognition evaluation value is used to indicate an average accuracy level of recognizing the target object in the target image.

[0113] In an exemplary embodiment, the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0114] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices. In an exemplary embodiment, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the present invention is not limited to any specific combination of hardware and software.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for determining an image recognition evaluation value, characterized in that: include: Obtain target quality parameters corresponding to the target image; Determining a plurality of identification reference values ​​corresponding to the target quality parameter according to a posterior probability model and a plurality of reference images; wherein the identification reference value is used to indicate the accuracy of identifying the target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object; Determining a target recognition evaluation value of the target image according to the multiple recognition reference values, wherein the target recognition evaluation value is used to indicate an average accuracy of recognizing a target object in the target image; The target quality parameter includes at least one of the following information: the image quality level of the target image, the posture level of the target object contained in the target image, and the completeness of the target object; Wherein, obtaining the target quality parameter corresponding to the target image includes: obtaining a plurality of reference images in a database, wherein the reference image and the target image contain the same target object; combining the target image and the plurality of reference images into a plurality of sets, wherein the quality parameters of the plurality of sets are the same, and the sets include: the target image and part of the reference images; obtaining the quality parameter of any set in the plurality of sets, and using the quality parameter of any set as the target quality parameter of the target image; Among them, determining multiple identification reference values ​​corresponding to the target quality parameters according to the posterior probability model and multiple reference images includes: inputting the multiple sets into the posterior probability model respectively according to a preset order; instructing the posterior probability model to determine the identification reference value corresponding to each set in the multiple sets according to the quality parameters of the multiple sets, so as to obtain multiple identification reference values, and using the multiple identification reference values ​​as the multiple identification reference values ​​corresponding to the target quality parameters.

2. The method for determining an image recognition evaluation value according to claim 1, characterized in that: Before determining a plurality of identification reference values ​​corresponding to the target quality parameter according to the posterior probability model and a plurality of reference images, the method further comprises: constructing a first set according to a first reference image and a first test image corresponding to the first reference image, and determining a first quality parameter of the first set, wherein a first target object of the first reference image is the same as a second target object of the first test image; Determining a matching degree between a first target object in the first reference image and a second target object in the first test image by a pedestrian re-identification algorithm; A first identification reference value of the first quality parameter is determined according to the matching degree.

3. The method for determining an image recognition evaluation value according to claim 2, characterized in that: After determining the first identification reference value of the first quality parameter according to the matching degree, the method further includes: The Gaussian mixture model is constructed using the following formula: in, μ k It is used to indicate the mean of the kth component of the Gaussian mixture model, Σ k Used to indicate the variance of the kth component of the Gaussian mixture model, q is the first quality parameter, r is the first identification reference value, wherein the Gaussian mixture model is used to indicate the probability distribution of the first quality parameter and the first identification reference value.

4. The method for determining an image recognition evaluation value according to claim 3, characterized in that: After constructing the Gaussian mixture model by the following formula, the method further includes: Solving the Gaussian mixture model by using a preset algorithm; The posterior probability model of the identification reference value is determined according to the solved Gaussian mixture model, wherein the posterior probability model is:

5. A device for determining an image recognition evaluation value, characterized in that: include: An acquisition module, used for acquiring target quality parameters corresponding to a target image; A first determination module is used to determine a plurality of identification reference values ​​corresponding to the target quality parameter according to a posterior probability model and a plurality of reference images; wherein the identification reference value is used to indicate the accuracy of identifying the target object in the target image based on the corresponding reference image and the target quality parameter, and the reference image and the target image contain the same target object; A second determination module, configured to determine a target recognition evaluation value of the target image according to the plurality of recognition reference values, wherein the target recognition evaluation value is used to indicate an average accuracy of recognizing a target object in the target image; The target quality parameter includes at least one of the following information: the image quality level of the target image, the posture level of the target object contained in the target image, and the completeness of the target object; The acquisition module is further used to acquire multiple reference images in a database, wherein the reference images and the target image contain the same target object; combine the target image and the multiple reference images into multiple sets, wherein the multiple sets have the same quality parameters, and the sets include: the target image and part of the reference images; acquire the quality parameter of any set in the multiple sets, and use the quality parameter of any set as the target quality parameter of the target image; Among them, the first determination module is also used to input the multiple sets into the posterior probability model respectively according to a preset order; instruct the posterior probability model to determine the identification reference value corresponding to each set in the multiple sets according to the quality parameters of the multiple sets, so as to obtain multiple identification reference values, and use the multiple identification reference values ​​as the multiple identification reference values ​​corresponding to the target quality parameters.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method described in any one of claims 1 to 4 when executed.

7. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 4 through the computer program.

Images