A small sample recognition method and system

Abstract

The application provides a small sample recognition method, comprising the following steps: S1, sample preparation, wherein samples are screened through a sample input module and a sample screening module; S2, model area division of the samples is performed through a computer analysis module; S3, according to the model area divided in the step S2, model area features are calculated, and a face is recognized through an image recognition module; and S4, region recognition is performed on a target image to be recognized, and recognition data is output. The application solves the problem that a large number of samples are required during image recognition, and the feature extraction problem can be solved through a small number of samples. Reasonable explanations can be given for the occurrence of recognition rejection and misrecognition. Since a large number of samples are not required and complex feature extraction is not required, the algorithm complexity is greatly reduced, and the cost of computing hardware is reduced.

Description

Technical Field

[0001] This invention relates to the field of sample recognition, and more specifically, to a method and system for small sample recognition. Background Technology

[0002] Modern society is an intelligent society. Pattern recognition technology can replace manual labor in many identification tasks, reducing labor intensity and difficulty. From video recognition in public security and crime prevention to banknote recognition in the financial industry, and image recognition in hospitals such as CT scans, pattern recognition technology is used in all of them.

[0003] The current mainstream pattern recognition technology is deep learning. Deep learning requires a large number of samples; typically, at least a thousand samples are needed for each category. This greatly limits the application scenarios of deep learning. Furthermore, deep learning algorithms are highly complex and require accelerator cards such as GPUs to achieve fast computation and real-time recognition. In addition, deep learning often faces the problem of uneven sample distribution, which can lead to non-convergence. Since feature extraction in deep learning is also done automatically, it is usually impossible to provide a reasonable explanation for cases of rejection and false recognition.

[0004] Existing deep learning requires a large number of samples and is computationally intensive, necessitating accelerator cards such as GPUs. Because it cannot provide an intuitive and reasonable explanation for feature extraction, it cannot provide a reasonable explanation for rejection and false recognition, nor can it effectively control rejection and false recognition to meet the industry's refined management requirements.

[0005] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a small sample identification method and system, which solves the problem of having to select a large sample for screening, typically requiring no more than ten samples. It also solves the problem of feature extraction and provides reasonable explanations for rejection and false recognition.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A small sample identification method includes...

[0009] S1: Sample preparation, which involves screening samples through the sample entry module and the sample screening module;

[0010] S2: The computer analysis module divides the sample into model regions;

[0011] S3: Based on the model region defined in step S2, calculate the features of the model region and recognize the face through the image recognition module;

[0012] S4: Perform region recognition on the target image to be recognized and output the recognition data.

[0013] As a preferred technical solution in this application, the method for dividing the model region in step S2 is as follows:

[0014] S11: Let the left coordinate of the X-axis of the object's partition region be Xl, the right coordinate be Xr; the lower coordinate of the Y-axis be Yb, the upper coordinate be Yt, and the pixel value of the face image be Pxy.

[0015] S12: The average pixel value in this area is:

[0016] S13: The pixel change measure for this region is:

[0017] S14: Divide the model region according to the average pixel value of the region and the pixel change metric of the region.

[0018] As a preferred technical solution of this application, the above-mentioned area is slid up, down, left and right and the area is increased or decreased to find the optimal area. The criterion for judging the optimal area is to select the largest pixel change measure in the area, which is denoted as Vm_max.

[0019] As a preferred technical solution of this application, the region pixel change metric Vm is the largest quantity in each candidate region. The left coordinate of the X-axis of this region is Xl_max, the right coordinate is Xr_max, the lower coordinate of the Y-axis is Yb_max, and the upper coordinate is Yt_max.

[0020] As a preferred technical solution of this application, the pixels of the target image region are denoted as Pxy_t, and the mean value is calculated by the formula for the mean value of region pixels and is denoted as Am_t.

[0021] As a preferred technical solution of this application, the similarity measurement formula between the target image region to be identified and the best region of the sample image is as follows:

[0022]

[0023] The Sem value represents the similarity between the image to be identified and the sample.

[0024] The pixel values ​​of the sample image region are denoted as Pxy_s, and the mean value is calculated using the formula for the mean pixel value of the region, denoted as Am_s. The pixel change measure of the target image region is Vm_t, and the pixel change measure of the sample image and target image regions is Vm_s. √ This refers to the square root of a mathematical formula.

[0025] As a preferred technical solution of this application, the sample model region division includes eyebrows, eyes, nose, mouth and micro-expressions.

[0026] A small sample recognition system includes a sample input module, a sample screening module, a sample model region division module, a camera module, an image recognition module, a computer analysis module, a recognition data output module, a power supply module, and a storage module, wherein the above modules are interconnected.

[0027] As a preferred technical solution of this application, the number of samples prepared does not exceed 10.

[0028] As a preferred technical solution of this application, the image recognition module and the computer analysis module are used for model region division and image recognition.

[0029] The working principle of this invention is as follows: Samples are selected through the sample screening module, and then no more than 10 samples are entered through the sample input module. The samples are divided into sample model regions, and images such as eyebrows, corners of the mouth, and nose are selected for image recognition. The sample model regions and image recognition are calculated by the computer analysis module, and the results are output through the data output module. The camera module records the recognized images and transmits them to the storage module for future use.

[0030] Compared with the prior art, the present invention has the following beneficial effects: (1) Compared with deep learning, the process of the present invention is greatly simplified and the amount of computation is greatly reduced; (2) The algorithm provided by the present invention can accurately distinguish the problem of uneven lighting on the left and right faces; (3) The feature extraction of the present invention is regularized and displayed, which is convenient for visualization and intuitive understanding; (4) The present invention can better handle the problems of rejection and false recognition, and better control the rejection rate and false recognition rate; (5) The present invention has the function of storing faces, which can be used when necessary; (6) The present invention does not require a large number of samples or complex feature extraction, the algorithm complexity is greatly reduced, and the cost of computing hardware is reduced.

[0031] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a sample view of the present invention;

[0034] Figure 2 This is a view of one embodiment of the present invention;

[0035] Figure 3 This is a view of another embodiment of the present invention;

[0036] Figure 4 This is a view of another embodiment of the present invention;

[0037] Figure 5 This is a view of another embodiment of the present invention.

[0038] In the picture: 1. Eyebrows; 2. Nose; 3. Eyes; 4. Corners of the mouth. Detailed Implementation

[0039] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses consistent with some aspects of this disclosure as detailed in the appended claims.

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0041] Example 1:

[0042] like Figure 1-2 As shown, first select a sample; when selecting to identify eyebrows:

[0043] 1. Calculate the mean and the change:

[0044] Let Xl be the left coordinate of the eyebrow division region, Xr be the right coordinate, Yb be the bottom coordinate of the Y axis, Yt be the top coordinate, and Pxy be the pixel value of the face image.

[0045] / ((Xr-Xl)(Yt-Yb))

[0046] 2. Find the optimal regional features:

[0047] To find the best feature of the best region, we need to slide the eyebrow area up, down, left, and right and increase or decrease the area to find the best region. The criterion for judging the best region is the pixel change measure Vm of the region. The region with the largest Vm among these candidate regions is selected and denoted as Vm_max.

[0048] The left coordinate of the X-axis of this region is Xl_max, and the right coordinate is Xr_max; the lower coordinate of the Y-axis is Yb_max, and the upper coordinate is Yt_max.

[0049] Select ten samples and process them in the same way. After the above regional division, calculate the ten optimal regions for the ten samples at the eyebrows.

[0050] 3. Region identification:

[0051] A target image of the eyebrow to be identified is set, and region recognition is performed. The region recognition process is as follows:

[0052] The pixel region of the target image is denoted as: Pxy_t

[0053] The mean is calculated using the pixel mean formula described above, and is denoted as Am_t.

[0054] The similarity metric formula between the target image's region to be identified and the optimal region of the sample image is as follows:

[0055]

[0056] Among the ten selected samples, the largest Sem value is chosen as the similarity between the image to be identified and the sample, and the image comparison result is output through the data output module.

[0057] According to the similarity formula, the maximum value of this similarity is 1, which means that the target to be identified is the sample itself. Generally, it is less than 1. The larger the value, the higher the similarity with the sample. Based on the meaning of similarity and the requirements of application scenarios for rejection and false recognition, a similarity threshold is set to solve the current problem of rejection and false recognition of samples in deep learning.

[0058] Example 2:

[0059] like Figure 3 As shown, first select a sample; when selecting to identify the nose:

[0060] 1. Calculate the mean and the change:

[0061] Let Xl be the left coordinate of the nose region on the X-axis and Xr be the right coordinate; Yb be the bottom coordinate of the Y-axis and Yt be the top coordinate; and Pxy be the pixel value of the face image.

[0062] / ((Xr-Xl)(Yt-Yb))

[0063] 2. Find the optimal regional features:

[0064] To find the best feature of the best region, we need to slide the nose region up, down, left and right and increase or decrease the region to find the best region. The criterion for judging the best region is the pixel change measure Vm of the region. The region with the largest Vm among these candidate regions is selected and denoted as Vm_max.

[0065] The left coordinate of the X-axis of this region is Xl_max, and the right coordinate is Xr_max; the lower coordinate of the Y-axis is Yb_max, and the upper coordinate is Yt_max.

[0066] Ten samples were selected and processed in the same way. After the above regional division, the ten optimal regions on the nose of the ten samples were calculated.

[0067] 3. Region identification:

[0068] A target image of the nose to be identified is set, and region recognition is performed. The region recognition process is as follows:

[0069] The pixel region of the target image is denoted as: Pxy_t

[0070] The mean is calculated using the pixel mean formula described above, and is denoted as Am_t.

[0071] The similarity metric formula between the target image's region to be identified and the optimal region of the sample image is as follows:

[0072]

[0073] Among the ten selected samples, the largest Sem value is chosen as the similarity between the image to be identified and the sample, and the image comparison result is output through the data output module.

[0074] According to the similarity formula, the maximum value of this similarity is 1, which means that the target to be identified is the sample itself. Generally, it is less than 1. The larger the value, the higher the similarity with the sample. Based on the meaning of similarity and the requirements of application scenarios for rejection and false recognition, a similarity threshold is set to solve the current problem of rejection and false recognition of samples in deep learning.

[0075] Example 3:

[0076] like Figure 4 As shown, first select a sample; when selecting to recognize eyes:

[0077] 1. Calculate the mean and the amount of change: Using the same method as in Examples 1 and 2, the eye is divided into regions, and the best region that the eye can recognize is analyzed by the computer analysis module;

[0078] 2. Find the optimal regional features:

[0079] Using the same method as in Examples 1 and 2, the best regional features were selected from the samples. Ten samples were selected and processed in the same way. After the above-mentioned regional division, the ten best regions at the eye of the ten samples were calculated.

[0080] 3. Region identification:

[0081] The target image is set for the eye to be identified, the image recognition module is used to identify the region, and the image comparison result is output through the data output module to obtain the similarity of the sample identification.

[0082] According to the similarity formula, the maximum value of this similarity is 1, which means that the target to be identified is the sample itself. Generally, it is less than 1. The larger the value, the higher the similarity with the sample. Based on the meaning of similarity and the requirements of application scenarios for rejection and false recognition, a similarity threshold is set to solve the current problem of rejection and false recognition of samples in deep learning.

[0083] Example 4:

[0084] like Figure 5 As shown, first select a sample. When selecting to recognize the corner of the mouth:

[0085] 1. Calculate the mean and the amount of change: Using the same method as in Examples 1 and 2, the corner of the mouth is divided into regions, and the best region that can be identified by the corner of the mouth is analyzed by the computer analysis module;

[0086] 2. Find the optimal regional features:

[0087] Using the same method as in Examples 1 and 2, the best regional features were selected from the samples. Ten samples were selected and processed in the same way. After the above-mentioned regional division, the ten best regions at the corners of the mouth of the ten samples were calculated.

[0088] 3. Region identification:

[0089] The target image of the corner of the mouth is set, the image recognition module is used to identify the region, and the image comparison result is output through the data output module to obtain the similarity of the sample recognition.

[0090] According to the similarity formula, the maximum value of this similarity is 1, which means that the target to be identified is the sample itself. Generally, it is less than 1. The larger the value, the higher the similarity with the sample. Based on the meaning of similarity and the requirements of application scenarios for rejection and false recognition, a similarity threshold is set to solve the current problem of rejection and false recognition of samples in deep learning.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0092] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope of this disclosure is indicated by the following claims.

Claims

1. A small sample recognition method, characterized in that, The method comprises the following steps: S1: sample preparation, screening the sample through a sample input module and a sample screening module; S2: dividing the model region of the sample through a computer analysis module; S3: calculating the model region characteristics according to the model region divided in step S2, and identifying the face through an image recognition module; S4: performing region identification on the target image to be identified, and outputting the identification data; The method for dividing the model region in step S2 is as follows: S11: setting the left coordinate of the X-axis of the object division region as Xl, the right coordinate of the X-axis as Xr, the lower coordinate of the Y-axis as Yb, and the upper coordinate of the Y-axis as Yt, and the pixel value of the face image as Pxy; S12: The average pixel value of this region is: Am=( ) / ((Xr-Xl)(Yt-Yb)); S13: The regional pixel variation metric is: Vm= ∑(Xr- Xl) * (Yt- Yb) / ((Xr- Xl)(Yt- Yb)); ) / ((Xr-Xl)(Yt-Yb)); S14: dividing the model region according to the pixel mean value and the region pixel variation measure; The above region is slid up, down, left and right, and the region is increased or decreased to find the best region, and the judgment standard of the best region is that the maximum is selected in the region pixel variation measure, which is recorded as Vm_max; The region pixel variation measure Vm is the maximum in each candidate region, the left coordinate of the X-axis of the region is Xl_max, the right coordinate of the X-axis is Xr_max, the lower coordinate of the Y-axis is Yb_max, and the upper coordinate of the Y-axis is Yt_max.

2. The small sample identification method of claim 1, wherein: The target image region pixel is recorded as: The mean value is calculated by the region pixel mean value formula and recorded as Am_t.

3. The small sample identification method of claim 1, wherein: The similarity measure formula of the target image to be identified region and the sample image best region is as follows: Sem= ( ) / ((Xr-Xl)(Yt-Yb)) / ((Xr-Xl)(Yt-Yb)) Sem value, similarity of the image to be identified and the sample; wherein the sample image region pixels are denoted as: s, the mean is calculated by the region pixel mean formula, denoted as: Am_s, the target image region pixel variation measure is: Vm_t, the sample image target image region pixel variation measure is: Vm_s, wherein is the square root of the mathematical formula.

4. The small sample identification method of claim 1, wherein: The sample model region division includes eyebrows, eyes, nose, mouth and micro-expression.

5. A small sample identification system using the identification method of any one of claims 1-4, characterized in that: The system comprises a sample input module, a sample screening module, a sample model region division module, a camera module, an image recognition module, a computer analysis module, an identification data output module, a power module and a storage module, and the above modules are electrically connected.

6. The small sample identification system of claim 5, wherein, The number of sample preparations is not more than 10.

7. The small sample identification system of claim 5, wherein The image recognition module and the computer analysis module are used for model region division and image recognition.

Images