A pattern recognition system captured by a camera

Abstract

The present invention relates to the field of image processing technology, and specifically to a pattern recognition system taken by a camera, comprising a priori layer, a recognition layer and a determination layer; an object image is uploaded through the priori layer, and the identification and marking of the object image is completed in the priori layer. After the object image is identified and marked, the storage operation of the object image is further performed, and the recognition layer drives the camera to run and collect images in real time. Based on the complexity of the collected image, the segmentation logic of the collected image is configured, and after the collected image is segmented based on the segmentation logic, the present invention can perform object recognition on the collected image, and in the process of recognizing objects in the image, it is no longer limited by the number and type of objects in the image. The overall robustness of the system is strong, the performance is relatively good, and the collected image can be well processed for object recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a pattern recognition system photographed by a camera. Background Art

[0002] Photo recognition refers to the technology of analyzing and identifying taken photos by using computer vision and image recognition technology to determine the objects or scenes contained in the photos. It can help people quickly and accurately obtain information about objects or scenes.

[0003] The invention patent with application number 202210107916.1 discloses an image recognition system based on video shooting, which is characterized by including an image acquisition control module, an image extraction module, a flash data analysis module, a ratio retrieval module and a target measurement module; the image acquisition control module is used to collect instantaneous image information at each time interval before and after the flash: the image extraction module is used to identify and extract the instantaneous image of the target before the flash and the instantaneous image of the flash reaching the target; the flash data analysis module is used according to the distance information of the flash reaching the target; the ratio retrieval module is used to retrieve the ratio adjustment factor corresponding to the flash distance; the target measurement module is used to calculate the target size data according to the captured image of the selected target. The image acquisition control module includes a flash signal acquisition module and an instantaneous image acquisition module: the flash signal acquisition module is used to collect the start and stop signals of the flash; the instantaneous image acquisition module is used to sequentially collect all instantaneous images during the flash start and stop process.

[0004] The application aims to solve the problem that "the actual size of the target object is obtained based on a reference object, which needs to be performed on the same plane, but it is impossible to accurately measure in different dimensions, and it is impossible to complete the size measurement of the target without a reference object."

[0005] However, current camera-based object recognition technology often focuses on the camera-based recognition of a single or several targets. When there are many targets in the captured image data, it is often impossible to comprehensively recognize all targets in the image data, resulting in the recognition result of only one or several targets. In order to identify all objects in the image data, the image data needs to be decomposed and photographed, that is, all targets contained in the original image data need to be covered through multiple image data collections. This process is relatively cumbersome and inefficient.

[0006] System Contents

[0007] In view of the above shortcomings of the prior art, the present system provides a pattern recognition system photographed by a camera, which solves the technical problems raised in the above background technology.

[0008] To achieve the above objectives, this system is implemented through the following technical solutions:

[0009] A camera-photographed pattern recognition system comprises a priori layer, a recognition layer and a determination layer;

[0010] The object image is uploaded through the prior layer, and the object image is marked in the prior layer. After the object image is marked, the object image storage operation is further performed. The recognition layer drives the camera to collect images in real time. Based on the complexity of the collected image, the segmentation logic of the collected image is configured. After the collected image is segmented based on the segmentation logic, the segmented image blocks are further used to search for suspected items in the object image stored in the prior layer. The determination layer receives the suspected item search results in the recognition layer in real time, and performs object determination on the image collected by the recognition layer based on the suspected item search results.

[0011] The recognition layer includes an acquisition module, an analysis module and a search module. The acquisition module is used to drive the camera to run and acquire images. The analysis module is used to receive the acquired images in the acquisition module, perform complexity analysis on the acquired images, configure segmentation logic for the acquired images based on the complexity analysis results of the acquired images, and apply the segmentation logic to segment the acquired images. The search module is used to receive the image blocks of the acquired images obtained by segmentation in the analysis module, and use the image blocks as search targets to search for suspected items in the object images stored in the priori layer.

[0012] The complexity of the collected image is calculated by the following formula:

[0013]

[0014] is the complexity of the distribution of the kth color in space; p k is the total number of pixels of the kth color in the image; F is the total number of pixels in the image;

[0015] The larger the value of the image complexity N is, the more objects or types there are in the image. u is the number of different connected domains of this color; nv is the number of pixels in the vth connected domain; S is the total number of pixels of this color.

[0016] Furthermore, the a priori layer includes an uploading module, a marking module and a storage module, the uploading module is used to upload the object image, the marking module is used to obtain the object image uploaded in the uploading module, mark the object image, and the storage module is used to receive the object image marked in the marking module and store the object image;

[0017] Among them, the uploading operation of the object image in the uploading module and the operation of marking the object image in the marking module are completed manually by the system end user, and the identification mark content of the object image is a text mark of the object name. During the operation stage of the storage module, when the object image that has completed the identification mark is stored, the identification mark of the object image is synchronously identified, and the object images with the same identification mark are stored in the same position, so that the object images with different identification marks are stored separately, and the object images uploaded in the uploading module only contain one group of objects.

[0018] Furthermore, during the operation phase of the upload module, the operation status of the marking module is synchronously monitored until the marking module completes the acquisition of all uploaded object images in the upload module, and then the upload module performs the object image upload operation again. The object images uploaded in the upload module are manually uploaded by system end users and images collected in the recognition layer that have not been judged by the judgment layer.

[0019] Among them, the images collected in the recognition layer but not completed by the determination layer are stored in the determination layer. When the upload module runs to upload the object image, the images collected in the recognition layer but not completed by the determination layer are sent to the upload module synchronously.

[0020] Furthermore, after the acquisition module drives the camera to acquire images, it simultaneously performs quality assessment on the acquired images, and performs noise reduction processing on the acquired images based on the quality assessment results. The quality assessment logic of the acquired images is expressed as follows:

[0021]

[0022] Where: PSNR is the peak signal-to-noise ratio of the image; MAX I is the maximum value of the pixel color in the image; m and n are the length and width of the image; I(i,j) is the pixel value of the pixel I at the position (i,j) in the image; K(i,j) is the pixel value of the pixel K adjacent to the pixel at the position (i,j) in the image:

[0023] Among them, if PSNR≥40db, it means that the quality of the collected image is good and no noise reduction processing is required. Otherwise, noise reduction processing is performed on the collected image.

[0024] Furthermore, the image color entropy ce is obtained by the following formula:

[0025]

[0026] Where: Q is the total color level after image quantization; x a is the proportion of pixels of level a color in the entire image. After the image color entropy ce is obtained, it is further normalized. The normalization logic is expressed as CE is the color entropy of the image after normalization; ce max is the maximum color entropy value in the image;

[0027] Among them, after the image color entropy ce is normalized, the CE obtained by the normalization process is used to iteratively collect the ce used in the calculation formula of the complexity of the image.

[0028] Furthermore, the segmentation logic configured for the collected image in the analysis module is subject to the following conditions: the higher the complexity of the collected image, the more divided blocks the collected image has, and the length and width of each group of divided image blocks are equal, which is specifically expressed as:

[0029] g = N × δ;

[0030] Where: g is the number of image blocks segmented in the vertical and horizontal directions; N is the complexity of the image; δ is the segmentation coefficient;

[0031] in, is the impact factor, d is the number of times the acquired image is subjected to denoising, and the value of g is rounded up.

[0032] Furthermore, the suspected item search logic in the search module is expressed as:

[0033]

[0034] Where: Sim(f1,f2) is the suspected item judgment value; H is the dimension of the feature descriptor; f1[h] is the corner point feature value of image f1 in the h dimension of the feature descriptor; f2[h] is the corner point feature value of image f2 in the h dimension of the feature descriptor; r1·r2 is the dot product of the graphic feature vectors of image f1 and image f2; ||r1|| and r2|| are the norms of the image feature vectors corresponding to image f1 and image f2 respectively;

[0035] Among them, the larger the suspected item judgment value Sim(f1,f2), the higher the similarity between the two groups of images. The f1 and f2 in the suspected item judgment value Sim(f1,f2) correspond to the image blocks obtained by segmentation of the acquired image and the object images stored in the prior layer, respectively. After all the image blocks and the object images stored in the prior layer are used to obtain the suspected item judgment values ​​through the above formula, the image blocks are further continuously combined with the adjacent image blocks, and the combined image blocks are again combined with the object images stored in the prior layer to obtain the suspected item judgment values ​​through the above formula.

[0036] Furthermore, the determination layer includes a setting module, a determination module and a feedback module, the setting module is used to set a suspected item determination threshold, receive the suspected item determination value calculated between each image block obtained in the search module and the object image, the determination module is used to apply the suspected item determination threshold and compare the obtained suspected item determination value to determine whether the image block is a suspected item of the object image, and the feedback module is used to continuously receive image blocks of non-suspected items, and pack the non-suspected item image blocks and send them to the priori layer;

[0037] Among them, the judgment module determines that the suspected item judgment value is within the judgment threshold, that is, it determines that there is a suspected item in the object image stored in the prior layer of the image block, and simultaneously configures the text identification mark corresponding to a group of object images with the highest suspected item judgment value on the current image block. Otherwise, the feedback module is triggered to run.

[0038] Furthermore, when the non-suspected image blocks packaged and sent to the prior layer come from the same acquired image and the total amount of the non-suspected image blocks exceeds one half of the size of the acquired image, the acquired image is used to iterate the corresponding non-suspected image blocks and send them to the prior layer.

[0039] Furthermore, the acquisition module is electrically connected to the analysis module and the search module through a medium, the search module is interactively connected to the storage module through a wireless network, the storage module is electrically connected to the marking module and the upload module through a medium, the acquisition module is interactively connected to the setting module through a wireless network, and the setting module is electrically connected to the determination module and the feedback module through a medium.

[0040] Compared with the known public technologies, the technical solution provided by this system has the following beneficial effects:

[0041] 1. The present invention provides a camera-photographed graphic recognition system. During operation, the system can perform object recognition on the captured image, and in the process of object recognition in the image, it is no longer limited by the number and type of objects in the image. The system has strong overall robustness and relatively good performance, and can perform object recognition processing on the captured image well.

[0042] 2. During the operation of the system of the present invention, at the image acquisition stage, the acquired images can be adaptively optimized based on their quality, so that when the acquired images are further processed by the system, subsequent image recognition processing can be completed with relatively stable image quality.

[0043] 3. In the present invention, the system, through the set logic, can synchronously complete the data collection of the object image for comparison during the operation of the system in the stage of object recognition processing of the collected image, so that the system has a certain degree of autonomy, making the operation of the system more intelligent, efficient and fast, and further improving the object recognition ability of the collected image with the increase of usage time. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present system or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present system, and those skilled in the art can obtain other drawings based on these drawings without creative work.

[0045] Figure 1 A schematic diagram of the structure of a pattern recognition system photographed by a camera;

[0046] Figure 2 A logical schematic diagram for collecting image segmentation and searching for suspected items in the present invention;

[0047] The numbers in the figure represent: A, acquired image; B, object image. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the embodiment of the present system clearer, the technical solution in the embodiment of the present system will be clearly and completely described below in conjunction with the drawings in the embodiment of the present system. Obviously, the described embodiment is a part of the embodiment of the present system, not all of the embodiments. Based on the embodiments in the present system, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present system.

[0049] The system is further described below in conjunction with embodiments.

[0050] Embodiment 1:

[0051] A pattern recognition system photographed by a camera in this embodiment, such as Figure 1 As shown, it includes a priori layer, recognition layer and decision layer;

[0052] The object image is uploaded through the prior layer, and the object image is marked in the prior layer. After the object image is marked, the object image storage operation is further performed. The recognition layer drives the camera to collect images in real time. Based on the complexity of the collected image, the segmentation logic of the collected image is configured. After the collected image is segmented based on the segmentation logic, the segmented image blocks are further used to search for suspected items in the object image stored in the prior layer. The determination layer receives the suspected item search results in the recognition layer in real time, and performs object determination on the image collected by the recognition layer based on the suspected item search results.

[0053] The recognition layer includes an acquisition module, an analysis module and a search module. The acquisition module is used to drive the camera to run and acquire images. The analysis module is used to receive the acquired images in the acquisition module, perform complexity analysis on the acquired images, configure segmentation logic for the acquired images based on the complexity analysis results of the acquired images, and apply the segmentation logic to segment the acquired images. The search module is used to receive the image blocks of the acquired images obtained by segmentation in the analysis module, and use the image blocks as search targets to search for suspected items in the object images stored in the priori layer.

[0054] The complexity of the acquired image is calculated using the following formula:

[0055]

[0056] Where: N is the complexity of the image; m and n are the length and width of the image; f(i,j) is the grayscale value of the pixel at the i-th row and j-th column in the image; is the grayscale average of all pixels in the 5×5 image block centered at (i, j); ce is the image color entropy; l is the set of pixel colors; Ck is the complexity of the kth color distribution in space; p k is the total number of pixels of the kth color in the image; F is the total number of pixels in the image;

[0057] The larger the value of the image complexity N is, the more objects or types there are in the image. u is the number of different connected domains of the color; n v is the number of pixels in the vth connected domain; S is the total number of pixels of this color;

[0058] The prior layer includes an upload module, a marking module and a storage module. The upload module is used to upload object images. The marking module is used to obtain the object images uploaded in the upload module and mark the object images. The storage module is used to receive the object images marked in the marking module and store the object images.

[0059] Among them, the upload operation of the object image in the upload module and the operation of marking the object image in the marking module are completed manually by the system end user, and the marking content of the object image is the text mark of the object name. During the operation stage of the storage module, when the object image with completed marking is stored, the marking of the object image is synchronously identified, and the object images with the same marking are stored in the same position, so that the object images with different markings are stored separately, and the object images uploaded in the upload module only contain one group of objects;

[0060] The image color entropy ce is obtained by the following formula:

[0061]

[0062] Where: Q is the total color level after image quantization; x a is the proportion of pixels of level a color in the entire image. After the image color entropy ce is obtained, it is further normalized. The normalization logic is expressed as CE is the color entropy of the image after normalization; ce max is the maximum color entropy value in the image;

[0063] Among them, after the image color entropy ce is normalized, the CE obtained by normalization is used to iteratively collect the ce used in the calculation formula of the complexity of the image;

[0064] The segmentation logic configured for the collected image in the analysis module follows the rule that the higher the complexity of the collected image, the more blocks the collected image will be segmented into, and the length and width of each group of segmented image blocks are equal, which can be specifically expressed as:

[0065] g = N × δ;

[0066] Where: g is the number of image blocks segmented in the vertical and horizontal directions; N is the complexity of the image; δ is the segmentation coefficient;

[0067] in, is the impact factor, d is the number of times the image is denoised, and the value of g is rounded up;

[0068] The logic of searching for suspected items in the search module is expressed as:

[0069]

[0070] Where: Sim(f1,f2) is the suspected item judgment value; H is the dimension of the feature descriptor; f1[h] is the corner point feature value of image f1 in the h dimension of the feature descriptor; f2[h] is the corner point feature value of image f2 in the h dimension of the feature descriptor; r1·r2 is the dot product of the graphic feature vectors of image f1 and image f2; ||r1|| and r2|| are the norms of the image feature vectors corresponding to image f1 and image f2 respectively;

[0071] Among them, the larger the suspected item determination value Sim(f1, f2), the higher the similarity between the two groups of images. The f1 and f2 in the suspected item determination value Sim(f1, f2) correspond to the image blocks obtained by segmenting the acquired image and the object images stored in the prior layer, respectively. After all the image blocks and the object images stored in the prior layer are used to obtain the suspected item determination value through the above formula, the image blocks are further combined with the adjacent image blocks continuously, and the image blocks obtained by the combination are again used with the object images stored in the prior layer to obtain the suspected item determination value through the above formula;

[0072] The determination layer includes a setting module, a determination module and a feedback module. The setting module is used to set a suspected item determination threshold, receive suspected item determination values ​​calculated between each image block obtained in the search module and the object image, the determination module is used to compare the suspected item determination threshold with the obtained suspected item determination value, and determine whether the image block is a suspected item of the object image. The feedback module is used to continuously receive image blocks of non-suspected items, and pack the non-suspected item image blocks and send them to the priori layer;

[0073] The determination module determines that the suspected item determination value is within the determination threshold, that is, determines that the image block has a suspected item in the object image stored in the prior layer, and simultaneously configures the text identification mark corresponding to a group of object images with the highest suspected item determination value on the current image block; otherwise, the feedback module is triggered to run;

[0074] The acquisition module is electrically connected to the analysis module and the search module through a medium, the search module is interactively connected to the storage module through a wireless network, the storage module is electrically connected to the marking module and the upload module through a medium, the acquisition module is interactively connected to the setting module through a wireless network, and the setting module is electrically connected to the determination module and the feedback module through a medium.

[0075] In this embodiment, the upload module is operated to upload the object image, the marking module obtains the object image uploaded in the upload module in real time, marks the object image, the storage module is operated to receive the object image marked in the marking module, and stores the object image, the acquisition module synchronously drives the camera to operate to acquire the image, the analysis module receives the acquired image in the acquisition module in real time, performs complexity analysis on the acquired image, configures the segmentation logic for the acquired image based on the complexity analysis result of the acquired image, and applies the segmentation logic to segment the acquired image, the search module further receives the image block of the acquired image obtained by the segmentation in the analysis module, and uses the image block as the search target to search for suspected items in the object image stored in the prior layer, and finally sets the suspected item determination threshold through the setting module, receives the suspected item determination value calculated between each image block obtained in the search module and the object image, the determination module runs to apply the suspected item determination threshold and compares it with the obtained suspected item determination value to determine whether the image block is a suspected item of the object image, and the feedback module finally continuously receives the image blocks of non-suspected items, packages the non-suspected item image blocks and sends them to the prior layer;

[0076] The above-mentioned calculation formula for the complexity of the acquired image provides necessary operation data support for the linkage operation of the recognition layer and the determination layer, making the operation of the determination layer more logical and ensuring that all object images in the acquired image can be recognized by the system;

[0077] See also Figure 2 As shown, based on the arrows in the figure, it can be known that the captured image A is segmented by segmentation logic to obtain several groups of image blocks, and further the image blocks obtained by segmentation and the image blocks obtained by further combining the segmented image blocks are used to perform suspected judgment with the object image B, and finally the object recognition operation of the captured image is completed.

[0078] Embodiment 2:

[0079] In terms of specific implementation, based on Example 1, this example refers to Figure 1 The following is a further detailed description of a pattern recognition system photographed by a camera in Example 1:

[0080] During the operation phase of the upload module, the operation status of the marking module is monitored synchronously until the marking module completes the acquisition of all uploaded object images in the upload module. The upload module then performs the object image upload operation again. The object images uploaded in the upload module are manually uploaded by system users and images collected in the recognition layer that have not been judged by the judgment layer.

[0081] Among them, the images collected in the recognition layer but not completed by the determination layer are stored in the determination layer. When the upload module runs to upload the object image, the images collected in the recognition layer but not completed by the determination layer are sent to the upload module synchronously.

[0082] Through the above settings, further operation logic is provided for the priori layer to ensure that the operation process of the priori layer is more stable.

[0083] like Figure 1 As shown, after the acquisition module drives the camera to collect images, it simultaneously performs quality assessment on the collected images, and performs noise reduction processing on the collected images based on the quality assessment results. The quality assessment logic of the collected images is expressed as:

[0084]

[0085] Where: PSNR is the peak signal-to-noise ratio of the image; MAX I is the maximum value of the pixel color in the image; m and n are the length and width of the image; I(i,j) is the pixel value of the pixel I at the position (i,j) in the image; K(i,j) is the pixel value of the pixel K adjacent to the pixel at the position (i,j) in the image:

[0086] Among them, if PSNR≥40db, it means that the quality of the collected image is good and no noise reduction processing is required. Otherwise, noise reduction processing is performed on the collected image.

[0087] Through the above settings, the quality of the images collected by the acquisition module in the recognition layer is evaluated, and the collected images are further denoised based on the quality evaluation results, so that the quality of the collected images applied by subsequent modules in the recognition layer is better, so as to facilitate further analysis and processing of the collected images.

[0088] like Figure 1 As shown, among the non-suspected item image blocks packaged and sent to the prior layer, when the non-suspected image blocks come from the same acquired image and the total amount of non-suspected image blocks exceeds one-half of the size of the acquired image, the acquired image is used to iterate the corresponding non-suspected image blocks and send them to the prior layer.

[0089] The above settings provide further operating logic for the feedback module, ensuring that the operation of the determination layer in the system can synchronously provide data support for the prior layer, thereby recycling the images collected by the system and further enhancing the performance of the object recognition function provided by the system.

[0090] In summary, during the operation of the system in the above embodiment, the system can identify objects in the captured images, and in the process of identifying objects in the images, it is no longer limited by the number and type of objects in the images. The system has strong overall robustness and relatively good performance, and can perform object recognition processing on the captured images well. Moreover, during the operation of the system, at the image acquisition stage, the captured images can be adaptively optimized based on their quality, so that when the captured images are further processed by the system, subsequent image recognition processing can be completed with relatively stable image quality. At the same time, the system, through the set logic in the stage of object recognition processing on the captured images, can synchronously complete the data collection of the comparison object images during the operation of the system, so that the system has a certain degree of autonomy, making the operation of the system more intelligent, efficient and fast, and further improving the system's ability to recognize objects in captured images as the use time increases.

[0091] The above embodiments are only used to illustrate the technical solutions of the present system, rather than to limit it. Although the present system has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present system.

Claims

1. A pattern recognition system photographed by a camera, characterized in that: It includes a priori layer, recognition layer and decision layer; The object image is uploaded through the prior layer, and the object image is marked in the prior layer. After the object image is marked, the object image storage operation is performed. The recognition layer drives the camera to run the acquisition image in real time. Based on the complexity of the acquired image, the segmentation logic of the acquired image is configured. After the acquired image is segmented based on the segmentation logic, the segmented image blocks are used to search for suspected items in the object image stored in the prior layer. The determination layer receives the suspected item search results in the recognition layer in real time, and performs object determination on the image acquired by the recognition layer based on the suspected item search results. The recognition layer includes an acquisition module, an analysis module and a search module. The acquisition module is used to drive the camera to run and acquire images. The analysis module is used to receive the acquired images in the acquisition module, perform complexity analysis on the acquired images, configure segmentation logic for the acquired images based on the complexity analysis results of the acquired images, and apply the segmentation logic to segment the acquired images. The search module is used to receive the image blocks of the acquired images obtained by segmentation in the analysis module, and use the image blocks as search targets to search for suspected items in the object images stored in the priori layer. The complexity of the acquired image is calculated by the following formula: Where: m, n are the length and width of the image; f(i, j) is the gray value of the pixel at the i-th row and j-th column in the image; is the grayscale average of all pixels in the 5×5 image block centered at (i, j); ce is the image color entropy; l is the set of pixel colors; Ck is the complexity of the kth color distribution in space; p k is the total number of pixels of the kth color in the image; F is the total number of pixels in the image; in, u is the number of different connected domains of the color; n v is the number of pixels in the vth connected domain; S is the total number of pixels of this color.

2. A camera-photographed pattern recognition system according to claim 1, characterized in that: The prior layer includes an upload module, a marking module and a storage module. The upload module is used to upload the object image. The marking module is used to obtain the object image uploaded in the upload module and mark the object image. The storage module is used to receive the object image marked in the marking module and store the object image. Among them, the uploading operation of the object image in the uploading module and the operation of marking the object image in the marking module are completed manually by the system end user, and the identification mark content of the object image is a text mark of the object name. During the operation stage of the storage module, when the object image that has completed the identification mark is stored, the identification mark of the object image is synchronously identified, and the object images with the same identification mark are stored in the same position, so that the object images with different identification marks are stored separately, and the object images uploaded in the uploading module only contain one group of objects.

3. A camera-photographed pattern recognition system according to claim 2, characterized in that: During the operation phase of the upload module, the operation status of the marking module is synchronously monitored until the marking module completes the acquisition of all uploaded object images in the upload module, and then the upload module performs the object image upload operation again. The object images uploaded in the upload module are manually uploaded by system end users and images collected in the recognition layer that have not been judged by the judgment layer. Among them, the images collected in the recognition layer but not completed by the determination layer are stored in the determination layer. When the upload module runs to upload the object image, the images collected in the recognition layer but not completed by the determination layer are sent to the upload module synchronously.

4. A camera-photographed pattern recognition system according to claim 1, characterized in that: After the acquisition module drives the camera to collect images, it simultaneously performs quality assessment on the collected images and performs noise reduction processing on the collected images based on the quality assessment results. The quality assessment logic of the collected images is expressed as follows: Where: PSNR is the peak signal-to-noise ratio of the image; MAX I is the maximum value of the pixel color in the image; m and n are the length and width of the image; I(i,j) is the pixel value of the pixel I at the position (i,j) in the image; K(i,j) is the pixel value of the pixel K adjacent to the pixel at the position (i,j) in the image: Among them, if PSNR≥40db, it means that the quality of the collected image is good and no noise reduction processing is required. Otherwise, noise reduction processing is performed on the collected image.

5. The image recognition system photographed by a camera according to claim 1, characterized in that: The image color entropy ce is obtained by the following formula: Where: Q is the total color level after image quantization; x a is the proportion of pixels of level a color in the entire image. After the image color entropy ce is obtained, normalization processing is further performed. The normalization processing logic is expressed as CE is the color entropy of the image after normalization; ce max is the maximum color entropy value in the image; Among them, after the image color entropy ce is normalized, the CE obtained by the normalization process is used to iteratively collect the ce used in the calculation formula of the complexity of the image.

6. A camera-photographed pattern recognition system according to claim 1, characterized in that: The segmentation logic configured for the collected image in the analysis module is subject to the following: the higher the complexity of the collected image, the more divided blocks the collected image has, and the length and width of each group of divided image blocks are equal, which is specifically expressed as: g = N × δ; Where: g is the number of image blocks segmented in the vertical and horizontal directions; N is the complexity of the image; δ is the segmentation coefficient; in, is the impact factor, d is the number of times the acquired image is subjected to denoising, and the value of g is rounded up.

7. The image recognition system photographed by a camera according to claim 1, characterized in that: The suspected item search logic in the search module is expressed as: Where: Sim(f1,f2) is the suspected item judgment value; H is the dimension of the feature descriptor; f1[h] is the corner point feature value of image f1 in the h dimension of the feature descriptor; f2[h] is the corner point feature value of image f2 in the h dimension of the feature descriptor; r1·r2 is the dot product of the graphic feature vectors of image f1 and image f2; ||r1|| and r2|| are the norms of the image feature vectors corresponding to image f1 and image f2 respectively; Among them, the larger the suspected item judgment value Sim(f1,f2), the higher the similarity between the two groups of images. The f1 and f2 in the suspected item judgment value Sim(f1,f2) correspond to the image blocks obtained by segmentation of the acquired image and the object images stored in the prior layer, respectively. After all the image blocks and the object images stored in the prior layer are used to obtain the suspected item judgment values ​​through the above formula, the image blocks are further continuously combined with the adjacent image blocks, and the combined image blocks are again combined with the object images stored in the prior layer to obtain the suspected item judgment values ​​through the above formula.

8. The image recognition system photographed by a camera according to claim 1, characterized in that: The determination layer includes a setting module, a determination module and a feedback module. The setting module is used to set a suspected item determination threshold, receive the suspected item determination value calculated between each image block obtained in the search module and the object image, the determination module is used to apply the suspected item determination threshold and compare the obtained suspected item determination value to determine whether the image block is a suspected item of the object image, and the feedback module is used to continuously receive image blocks of non-suspected items, and pack the non-suspected item image blocks and send them to the priori layer; Among them, the judgment module determines that the suspected item judgment value is within the judgment threshold, that is, it determines that there is a suspected item in the object image stored in the prior layer of the image block, and simultaneously configures the text identification mark corresponding to a group of object images with the highest suspected item judgment value on the current image block. Otherwise, the feedback module is triggered to run.

9. A camera-photographed pattern recognition system according to claim 8, characterized in that: When the non-suspected image blocks packaged and sent to the prior layer come from the same acquired image and the total amount of non-suspected image blocks exceeds one half of the size of the acquired image, the corresponding non-suspected image blocks are iterated with the acquired image and sent to the prior layer.

10. The image recognition system photographed by a camera according to claim 1, characterized in that: The acquisition module is electrically connected to the analysis module and the search module via a medium, the search module is interactively connected to the storage module via a wireless network, the storage module is electrically connected to the marking module and the upload module via a medium, the acquisition module is interactively connected to the setting module via a wireless network, and the setting module is electrically connected to the determination module and the feedback module via a medium.

Images