Single-sample Face Recognition Method Based on Multi-scale Dynamic Error Coding and Discriminant Probability Classification

Abstract

The present invention relates to a single-sample face recognition method based on multi-scale dynamic error coding and discriminant probability classification, belonging to the field of digital image and pattern recognition. The method comprises the following steps: S1: error extraction stage, in which reconstruction errors are obtained using extended sparse representation via a feature pyramid; S2: error coding stage, in which reconstruction errors at different scales are corrected via dynamic error coding, and multi-scale fusion is used to complete the face recognition process; S3: discriminant probability classification stage, in which errors of different blocks are combined and classified based on probability to obtain weight coefficients. The present invention extracts reconstruction errors from a multi-scale perspective and uses dynamic error coding for correction and fusion classification. This method can achieve better results and is more robust in single-sample face recognition.

Description

Technical Field

[0001] The present invention belongs to the field of digital image and pattern recognition, and relates to a single-sample face recognition method based on multi-scale dynamic error coding and discriminant probability classification. Background Art

[0002] Facial recognition technology is a highly accurate, low-intrusive biometric identification method widely used in information security, law enforcement and surveillance, smart cards, access control, and other fields. In real-world applications, such as ID card and passport recognition, judicial confirmation, and access control, only a small number of samples are often available. In these situations, traditional facial recognition methods cannot effectively identify faces with a small number of samples, requiring multiple facial images per subject to achieve good system performance. Furthermore, samples are often corrupted by interfering variables, and the small number of training samples makes it difficult to remove these interfering variations between training and test faces, making it difficult to achieve good recognition performance.

[0003] Existing methods face two difficulties. First, single-sample face recognition suffers from a limited number of samples, resulting in a lack of available information and degraded performance. Another issue is that existing methods are unable to handle occluded regions when dealing with unknown variations. Therefore, the goal of this paper's approach is to fully utilize the discriminative information of a single sample by using both local and holistic methods, and to enable the dynamic error coding module to minimize the negative impact of interfering occluded local patches on the final classification. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a single-sample face recognition method based on multi-scale dynamic error coding and discriminant probability classification to solve the face recognition problem in real life where each person has only one training image, and to improve the face recognition effect based on a single sample.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A single-sample face recognition method based on multi-scale dynamic error coding and discriminant probability classification includes the following steps:

[0007] S1: Error extraction stage: The reconstruction error is obtained by using extended sparse representation through feature pyramid;

[0008] S2: Error coding stage: Dynamic error coding is used to correct reconstruction errors at different scales, and multi-scale fusion is used to complete the face recognition process;

[0009] S3: Discriminant probability classification stage: The errors of different blocks are combined to obtain weight coefficients based on probability and classified.

[0010] Furthermore, the step S1 includes the following steps:

[0011] S11: Construct training dictionary X and intra-class variation dictionary v;

[0012] S12: Divide the training dictionary X, the intra-class variation dictionary V, and the test sample y into 1, 4, and 16 image blocks of the same size, respectively;

[0013] S13: Calculate the representation coefficient of each image block of the test sample y at each scale L. The calculation formula is:

[0014]

[0015] in, and To solve the sparse representation coefficient at the Lth scale, and satisfy the minimum value in formula (1); and Denote the training dictionary and intra-class variation dictionary of the jth block at the Lth scale, θL and represents the sparse representation coefficient at the Lth scale, μ is the regularization parameter;

[0016] S14: Solve the 1-norm minimization problem in step S13 using the Homotopy Method;

[0017] S15: Calculate the representation error of each image block j of the test sample y at each scale L The calculation formula is:

[0018]

[0019] Where i represents the i-th category, Represents the j-th test image at the L-th scale, function δ i Output a vector whose only non-zero columns are The column associated with the i-th category in .

[0020] Further, the step S2 includes the following steps:

[0021] S21: Calculate the weight coefficient w of the representation error of each block at each scale L (e ij ), the calculation formula is:

[0022]

[0023] in, α and β are adjustable parameters;

[0024] S22: Multiply the weight coefficient of each block by the error to obtain the corrected error The calculation formula is:

[0025]

[0026] Furthermore, step S3 specifically includes the following steps:

[0027] S31: The corrected error is subjected to a probability classification strategy to adjust the weight of the local block according to the credibility, thereby performing classification. The specific process is as follows:

[0028]

[0029]

[0030]

[0031]

[0032]

[0033] Among them, ζ(·) is a function arranged from small to large values, and Represents e j The smallest value and the second smallest value after sorting from small to large; sort represents a function that sorts from small to large;

[0034] S32: Calculate the credibility of the local block classification as the weight coefficient of the current block, and then re-obtain a new representation residual

[0035]

[0036]

[0037] Where exp represents the exponential function, and the parameters μ and δ are adjustable so that when When it is larger, is close to 1, and when Near 0 o'clock, Close to 0;

[0038] S33: Add up the representation errors of each block at each scale, and the category with the smallest error is the category of the test sample:

[0039]

[0040] The beneficial effects of the present invention are that, compared with extended sparse representation-based classification methods, the present method fully utilizes the discriminative information of a single sample using local and global approaches, and uses the dynamic error coding module to minimize the negative impact of interfering occluded local blocks on the final classification. The present invention extracts reconstruction errors from a multi-scale perspective and uses dynamic error coding for corrective fusion classification. Compared with other single-sample face recognition methods, the present method achieves better and more robust results in single-sample face recognition.

[0041] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, in which:

[0043] Figure 1 Flowchart of the single-sample face recognition method based on multi-scale dynamic error coding and discriminant probability classification. DETAILED DESCRIPTION

[0044] The following describes the embodiments of the present invention by means of specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0045] Among them, the accompanying drawings are only for illustrative purposes and represent only schematic diagrams rather than actual pictures, and should not be understood as limiting the present invention. In order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the dimensions of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions may be omitted in the accompanying drawings.

[0046] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "back", etc. indicating directions or positional relationships, they are based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operate in a specific direction. Therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0047] like Figure 1 As shown, the present invention provides a single-sample face recognition method based on multi-scale dynamic error coding and discriminant probability classification, which includes the following steps:

[0048] Step 1: Divide the training dictionary, intra-class variation dictionary and test samples into S image blocks, and then calculate the category residuals of different image blocks at different levels, where and are the jth block of the Lth scale of the training dictionary, the intra-class variation dictionary, and the test sample, respectively. Define the scale L∈[1, 2, 3]. When L=1, 2, 3, the total number of image blocks S=1, 4, 16.

[0049] Step 2: Use the following model to describe the sparsity problem of the j-th block at the L-th scale:

[0050]

[0051] Among them, θ and is the sparse representation coefficient, and μ is the regularization parameter.

[0052] Step 3: Use the Homotopy Method to solve the 1-norm minimization problem in Step 2.

[0053] Step 4: Calculate the representation error of each image block j of the test sample y at each scale L. The calculation formula is:

[0054]

[0055] where the function δ i Output a vector whose only non-zero column is The column associated with the i-th category in .

[0056] Step 5: Calculate the weight coefficient of the representation error of each block at each scale. The calculation formula is:

[0057]

[0058] in, α and β are adjustable parameters.

[0059] Step 6: Multiply the weight coefficient of each block by the error to obtain the corrected error. The calculation formula is:

[0060]

[0061] Step 7: The corrected error is passed through the discriminant probability classification strategy, which can be classified by adjusting the weight of the local block according to the credibility. The specific process is as follows:

[0062]

[0063]

[0064]

[0065]

[0066]

[0067] Among them, ζ(·) is a function arranged from small to large values, and Represents e j The smallest and second smallest values after sorting from smallest to largest.

[0068] Step 8: The credibility of the local block classification can be further calculated as the weight coefficient of the current block and then the new representation residual can be obtained. The calculation formula is:

[0069]

[0070]

[0071] Among them, the parameters μ and δ are adjustable so that when When it is larger, is close to 1, and when Near 0 o'clock, Close to 0.

[0072] Step 9: Add up the representation errors of each block at each scale, and the category with the smallest error is the category of the test sample.

[0073]

[0074] In order to verify the effect of the present invention, the following experiments were carried out:

[0075] The proposed method was tested for single-sample face recognition on the AR and CAS-PEAL databases. 80 categories were selected, with one natural face image from each category used for testing. Eight other images containing variations in lighting, expression, and occlusion were also used. The remaining 20 categories were used to construct a dictionary for intra-class variation. The experimental results are shown in Table 1.

[0076] Table 1

[0077]

[0078] On the CAS-PEAL database, we selected the first 150 categories for testing. Natural faces were used as training samples, and three images of people wearing glasses and three images of people wearing hats were grouped as test samples. The remaining 50 categories were used to construct a dictionary for intra-class variation. The experimental results are shown in Table 2.

[0079] Table 2

[0080]

[0081] From the experimental comparison results in the above two tables, it can be seen that the single-sample face recognition method designed in the present invention can improve the recognition rate and is robust to occlusion.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A single-sample face recognition method based on multi-scale dynamic error coding and discriminant probability classification, characterized by: The following steps are involved: S1: Error extraction stage: Reconstruction error is obtained by using extended sparse representation through feature pyramid. Step S1 includes the following steps: S11: Construct training dictionary X and intra-class variation dictionary V; S12: Divide the training dictionary X, the intra-class variation dictionary V, and the test sample y into 1, 4, and 16 image blocks of the same size, respectively; S13: Calculate the representation coefficient of each image block of the test sample y at each scale L. The calculation formula is: in, and To solve the sparse representation coefficient at the Lth scale, and satisfy the minimum value in formula (1); and denote the training dictionary and intra-class variation dictionary of the jth block at the Lth scale, θ L and represents the sparse representation coefficient at the Lth scale, μ is the regularization parameter; S14: Solve the 1-norm minimization problem of step S13 using the homology method; S15: Calculate the representation error of each image block j of the test sample y at each scale L The calculation formula is: Where i represents the i-th category, Represents the j-th test image at the L-th scale, function δ i Output a vector whose only non-zero columns are The columns associated with the i-th category in ; S2: Error coding stage: Correct the reconstruction errors of different scales through dynamic error coding, and complete the face recognition process through multi-scale fusion; step S2 includes the following steps: S21: Calculate the weight coefficient w of the representation error of each block at each scale L (e ij ), the calculation formula is: in, α and β are adjustable parameters; S22: Multiply the weight coefficient of each block by the error to obtain the corrected error The calculation formula is: S3: Discriminant probability classification stage: The errors of different blocks are combined to obtain weight coefficients based on probability and classified.

2. The single-sample face recognition method based on multi-scale dynamic error coding and discriminant probability classification according to claim 1, characterized in that: The step S3 specifically includes the following steps: S31: The corrected error is subjected to a probability classification strategy to adjust the weight of the local block according to the credibility, thereby performing classification. The specific process is as follows: in, It is a function that arranges values from small to large. and Represents e j The smallest value and the second smallest value after sorting from small to large; sort represents a function that sorts from small to large; S32: Calculate the credibility of the local block classification as the weight coefficient of the current block, and then re-obtain a new representation residual Where exp represents the exponential function, and the parameters μ and δ are adjustable so that when When it is larger, is close to 1, and when Near 0 o'clock, Close to 0; S33: Add up the representation errors of each block at each scale, and the category with the smallest error is the category of the test sample:

Images