WO2018151357A1 - Human face recognition method based on improved multi-channel cabor filter - Google Patents

Abstract

The present invention relates to a human face recognition method based on an improved multi-channel Gabor filter. The present invention comprises: a first step of acquiring a feature image when an input is of face images of N persons (N is a natural number of two or more) and an output is of an image of a stepped particle histogram feature vector expression of statistics; a second step of dividing the respective face images into a plurality of overlapping sub-images in the same manner; and a third step of generating a histogram of all the sub-images so as to extract a stepped histogram feature vector sequence. Therefore, the Gabor filter and center-symmetric local binary patterns (CS-LBP) are combined, thereby extracting feature points, which are robust against noise, and providing an effect of enabling high dimensionality of the extracted face feature point to be reduced.

Description

Enhanced Human Face Recognition Based on Multichannel Gabor Filters

The present invention relates to an improved multi-channel Gabor filter-based human face recognition method. More specifically, the Gabor filter and the CS-LBP in order to reduce the noise robust feature extraction and the high dimensionality of the extracted facial feature points The present invention relates to an improved multi-channel Gabor filter-based human face recognition method for human face recognition combining center-symmetric local binary patterns.

There is a growing interest in establishing a reliable access control system based on personal identification for a safer society. In order to more securely verify the identity, biometrics technology using human-specific biometric information is being actively researched rather than the conventional token method (card, key, etc.). Among the biometrics, face recognition has less user rejection and is the most natural biometric recognition method. Subsequently, a lot of research effort is concentrated. However, due to differences in the environment such as lighting, posture, facial expressions, and time, even the face image of the same person varies greatly, and in some cases, the correlation between the face images of other people is different from the correlation between images of the same person. Can be higher. For this reason, it is well known that it is very difficult to develop a stable face recognition algorithm irrespective of lighting, posture, facial expression, and time.

The present invention is to solve the above problems, by combining the Gabor filter and the center-symmetric local binary patterns (CS-LBP), to extract the robust feature point extraction and high dimensionality of the extracted facial feature point An object of the present invention is to provide an improved multi-faceted Gabor filter-based human face recognition method.

In addition, the present invention reduces the dimension of the feature image by combining Gabor feature images in different directions and scales, and improves human face recognition based on multi-channel Gabor filter for extracting low-dimensional facial feature points based on CS-LBP from feature images. It is to provide a method.

In addition, the present invention provides an improved multi-channel for achieving high accuracy of face recognition while reducing feature dimensions, storage space, and computation time through the Gabor filter and CS-LBP combining method, compared to the conventional Gabor filter and LBP approach. It is to provide a Gabor filter-based human face recognition method.

However, the objects of the present invention are not limited to the above-mentioned objects, and other objects not mentioned will be clearly understood by those skilled in the art from the following description.

In order to achieve the above object, an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention includes an input of N face (N is a natural number of 2 or more).

Output is an image of a cascading particle histogram feature vector representation of the statistics.

Feature image

(

,

A first step of obtaining; bracket

In the same manner, a second step of separating a plurality of overlapping sub-images; And

Generating a histogram of all sub-images of the, and extracting a cascading histogram feature vector sequence; Characterized in that it comprises a.

At this time, after the third step, respectively

Extracting a corresponding eigenvector of the cascade sequence of; It characterized in that it further comprises.

Also, after the fourth step, a fifth step of extracting a face image feature vector after the feature extraction of the first to fourth steps; Characterized in that further comprises,

In addition, the first step is a face image

Step 1-1 of obtaining a heirloom amplitude spectrum by performing heirloom conversion; For the amplitude spectrum of n different by overlap,

Obtaining a first step 1-2; And for each CS-LBP code, an image

Obtaining the first 1-3 steps; Characterized in that it comprises a.

In the improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention, a Gabor filter and a center-symmetric local binary patterns (CS-LBP) are combined to extract feature points that are robust to noise and extracted facial feature points. It provides the effect of reducing high dimensionality.

In addition, the improved multi-channel Gabor filter-based human face recognition method according to another embodiment of the present invention reduces the dimension of the feature image by combining the Gabor feature images in different directions and scales, and based on CS-LBP based low dimension from the feature image. Provides the effect of extracting facial feature points.

In addition, the improved multi-channel Gabor filter-based human face recognition method according to another embodiment of the present invention is characterized by a feature dimension, a storage space, and a combination of the Gabor filter and the CS-LBP method, compared to the conventional Gabor filter and LBP approach. It reduces the computation time and at the same time provides the effect of high accuracy of face recognition.

1 is a diagram illustrating a feature extraction algorithm module 100 in which an improved multi-channel Gabor filter-based human face recognition method is performed according to an embodiment of the present invention.

2 illustrates an image of a Gabor amplitude spectrum used in an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention.

3 is a diagram illustrating an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention.

4 is a diagram illustrating a Yale face database used in an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention.

FIG. 5 illustrates the effect of image block size and bin on CS-LBP for explaining an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention, and FIG. 6 illustrates image block size and storage for LBP. A graph showing the effect of.

FIG. 7 illustrates an ORL face database used in an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention.

FIG. 7 illustrates a FERET face database used in an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention.

Hereinafter, the detailed description of the preferred embodiment of the present invention will be described with reference to the accompanying drawings. In the following description of the present invention, detailed descriptions of well-known functions or configurations will be omitted when it is deemed that they may unnecessarily obscure the subject matter of the present invention.

In the present specification, when one component 'transmits' data or a signal to another component, the component may directly transmit the data or signal to another component, and through at least one other component. This means that data or signals can be transmitted to other components.

1 is a diagram illustrating a feature extraction algorithm module 100 in which an improved multi-channel Gabor filter-based human face recognition method is performed according to an embodiment of the present invention. Referring to FIG. 1, in the enhanced multi-channel Gabor filter-based human face recognition method, a "feature extraction algorithm module 100" based on a combination of Center-Symmetric Local Binary Patterns (CS-LBP) and Gabor Wavelet is performed. do. "Feature extraction algorithm module 100" is a Gabor filter using Gabor Wavelet transform means 110 and a CS-LBP operator to perform Gabor wavelet transform combined with CS-LBP face image feature extraction algorithm. Image texture extracting function means 120.

First, the Gabor Wavelet transform means 110 extracts a multi-scale image using optional attributes such as good spatial locality, spatial frequency and image.

Second, the Gabor filter image texture extraction function means 120 using the CS-LBP operator improves the robustness of the external environment change. This is because changes in the external environment, such as the direction of the region and other important features, lighting, expression, posture and shade, are robust.

Accordingly, when compared to the Gabor wavelet transform combined with the conventional LBP algorithm, the algorithm based on the combination of Center-Symmetric Local Binary Patterns (CS-LBP) and Gabor Wavelet of the present invention is an algorithm in space and time. In addition to reducing the overhead of, it can represent a significant recognition rate.

First, the Gabor filter image texture extracting function unit 120 using the CS-LBP operator may analyze the gray scale change of the scale and the image in all directions by a group of various scales having different filter directions.

The Gabor filter image texture extraction function means 120 using the CS-LBP operator has good time-frequency localization and multiple resolution characteristics, and has the ability to extract local nuances of an image. Thus, the Gabor filter image texture extraction function means 120 using the CS-LBP operator ensures robustness in lighting variations, image rotations and deformations. The Gabor wavelet kernel function used by the Gabor filter image texture extraction function means 120 using the CS-LBP operator is expressed by Equation 1 below.

here,

,

,

,

Since μ and v correspond to the direction and size of the filter, respectively, adjust the μ and v to select the direction and dimension of the filter. Where v ∈ {0, 1, 2, 3, 4}, μ ∈ {0, 1, 2, 3, 4, 5, 6, 7},

,

to be.

Meanwhile, FIG. 2 shows images of Gabor amplitude spectra used in an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention.

Referring to Fig. 2, an image of the Gabor amplitude spectrum is shown, and the relevant parameters are v 0 {0, 1, 2}, μ ∈ {0, 1, 2, 3, 4, 5, 6, 7}, σ = 2π,

,

to be.

The feature extraction algorithm module 100 combines Gabor filter images according to the following two methods to reduce the size of the Gabor function, and uses the multi-channel Gabor function to reduce the number of Gabor filter images and maintain the images. Extract the multiscale image from the original image. Here, Gabor filter images of the two combination methods are as follows.

First, the feature extraction algorithm module 100 adjusts each scale in the direction of the Gabor filter image overlay in various scales of the multi-channel Gabor function image to obtain a multi-frequency Gabor channel (MFGC).

Second, the feature extraction algorithm module 100 obtains a multi-directional Gabor channel (MOGC) in each direction in the scaled image overlay in the other direction of the multi-channel Gabor feature image Gabor filter.

3 is a diagram illustrating an improved multi-channel Gabor filter-based human face recognition method according to an embodiment of the present invention. Referring to FIG. 3, the improved multi-channel Gabor filter-based human face recognition method is performed by the feature extraction algorithm module 100 of FIG. 1, and the feature extraction algorithm module 100 includes a multi-channel Gabor filter and a CS-LBP. It may be performed by combining MOGC and CS-LBP based on facial image feature extraction algorithm.

In other words, the feature extraction algorithm module 100 may include image Gabor wavelet transform, superposition of Gabor amplitude spectra, superposition feature CS-LBP combination after MOGC extracted image coding, block and statistical histogram, for face feature extraction and face feature reduction. Cascade histogram style feature vector sequencing can be performed.

If the input to the feature extraction algorithm module 100 is a face image of N people (N is a natural number of 2 or more)

Output is an image of a cascading particle histogram feature vector representation of the statistics.

In the first step, the feature extraction algorithm module 100

Acquire feature images,

,

to be.

More specifically, the face image in step 1-1

Heirloom transformed to obtain a heirloom amplitude spectrum, and for superposed n amplitude spectra by superposition in steps 1-2,

And obtain an image for each CS-LBP code in steps 1-3.

Acquire.

Next, in the second step, the feature extraction algorithm module 100

In the same way, separate multiple overlapping sub-images.

Next, in the third step, the feature extraction algorithm module 100

Create a histogram of all sub-images of, and extract the cascading histogram feature vector sequence.

Next, in the fourth step, the feature extraction algorithm module 100

Extract the corresponding eigenvectors of the cascade sequence.

The feature extraction algorithm module 100 extracts a face image feature vector after feature extraction in the first to fourth steps described above.

Here, the feature extraction algorithm module 100 in the face recognition step of Equation 2

You can use the distance function to calculate test samples and learn sample similarities.

In Equation 2, T denotes a one-dimensional histogram feature vector of the training sample, S denotes a test sample one-dimensional histogram eigenvector, P denotes the number of sub-images, and Q denotes the number of sub-image histogram bins. r and i are exponents for P and Q, respectively. When characterizing the test sample and each training sample, the nearest neighbor classifier principle is a simple calculation that is widely used for face recognition, which improves the effectiveness of classifying test samples.

Hereinafter, the test results and analysis of the above-described improved multi-channel Gabor filter-based human face recognition method will be described. Here, we use the standard libraries for LBP, CS-LBP, Gabor + LBP based on MFGC and MOGC CS-LBP in Yale, ORL, and FERET to compare the recognition performance of the algorithm. The test assumes that all images are cropped and scaled to 64x64 using bilinear interpolation.

First, in parameter selection, the CS-LBP operator and the LBP operator for parameter selection are used. Blocking the size of sub-images affects recognition performance. If the block is too large, the extreme block size is the original image size and cannot reflect the benefits of local areas of image analysis.

The extreme situation of block image pixel level analysis is so small that it is easy to increase image registration sensitivity, increase computational complexity, and introduce the characteristics of noise images. At the same time, selecting the number of histogram bins had a certain effect on recognition performance. To properly select the block size and histogram bins, we use the Yale Face Standard Library to select the best suited for CS-LBP and LBP block image sizes and histogram bins.

Yale's face database has 11 images, including 15 people, and a total of 165 positive face images, including facial expressions and contrast changes. A portion of Yale is some sample image database of a person at Yale, such as FIG.

During the test, you randomly select five images and perform the remaining image tests, but repeat the test ten times.

The lower image size and histogram bin number are limited to test results for the effects of CS-LBP and LBP as shown in FIGS. 5 and 6, respectively. 5 shows the effect of image block size and bin on CS-LBP, and FIG. 6 shows the effect of image block size and storage on LBP.

As shown in FIG. 5, when the image block is not large, the number of bins on the recognition rate is not very large. Conversely, the smaller the number of bins, the lower the recognition rate. Regardless of whether the bin is 16, 8, or 4, the recognition rate is all 8 × 8 as the image block size, and when the bins are 16 and 8, the two recognition rate curves are very close, so we choose the CS-LBP algorithm. Where the parameter image block size is 8 * 8 and the number of bins is 8.

Referring to FIG. 6, when the image block size is 4 * 4, the number of bins on the recognition rate is not very large. When the number of bins is 256, the total optimal recognition rate curve and 8 * 8 when the image block size is the highest, so LBP at 8 * 8 and 256 bins for the LBP algorithm parameters, respectively. And CS-LBP algorithm are selected as the optimal parameters used in the test described below.

We look at the comparison of algorithm performance using these parameters.

Five feature dimensional algorithms are used to extract the five sample tests using the nearest neighbor classifiers and to train the sample time needed to compare the tests. The time required for training and testing is 10 times the test average. The test results are shown in Table 1, which shows the feature point dimensions and the training and test time results.

	Feature dimension	Training time (s)	Test time (s)
LBP	16384	14.30	0.06
CS-LBP	512	0.27	0.03
Gabor + LBP	373216	454.0	7.0
MFGC + CS-LBP	1536	12.32	0.11
MOGC + CS-LBP	4096	17.10	0.27

In Table 1, (1) the CS-LBP algorithm based on the MFGC and MOGC extracting the feature dimension is 1/32 of the LBP, and the CS-LBP algorithm extracts the feature point dimension much lower than the LBP algorithm. ) CS-LBP has a greater advantage in the time required for training samples, (3) the CS-LBP algorithm is still superior in terms of test time, and takes half of LBP. The longest Gabor + LBP algorithm requires 7.0 seconds and the MOGC + CS-LBP algorithm requires 0.27 seconds. Therefore, CSP-LBP extracts feature point dimensions over LBP, has more advantages in the time required for training and test samples, and can perform image feature point extraction more effectively.

Next, if you look at the test results of Yale's face database, the Yale's face database was introduced in detail in the face of the library for the basic situation, so it is omitted here. The test plan randomly selects 3, 4, 5, 6, 7 images as training samples, the rest as test samples, and repeats 10 tests. Table 2 shows the results of the algorithm's recognition rate on Yale's face database according to five kinds of test comparison algorithms.

Training sample number	3	4	5	6	7
LBP	88.17	89.33	90.78	92.40	93.00
CS-LBP	88.83	90.67	92.33	93.33	93.50
Gabor + LBP	89.45	91.33	93.20	94.07	95.00
MFGC + CS-LBP	85.50	88.19	90.56	93.00	93.83
MOGC + CS-LBP	90.08	91.43	93.22	94.27	95.00

As shown in Table 2, (1) the CS-LBP algorithm achieved a significant recognition rate compared to the LBP algorithm, and the recognition rate increased by almost 1% for 4 and 6 in the test sample. (2) Gabor + LBP algorithm, MOGC + CS-LBP algorithm achieved a significant recognition rate compared to the LBP algorithm, and the recognition rate improved about 2% compared to the CS-LBP algorithm.

Next, look at the test results of the ORL face database, the ORL face database contains 40 people, all have 10 images, a total of 400 face images. Among the 10 ORL face image databases of the face images, some deflections, blocks, and the like are included in each individual image as shown in FIG. 7. That is, FIG. 7 shows ten sample images of a person in an ORL. In the test, each of 3, 4, 5, and 6 images was randomly selected as the training sample and the rest of the images were randomly selected as the test sample. Table 3 shows the test results for the five types of algorithms in the iterative test. That is, Table 3 shows the recognition rate results of the five algorithms for the ORL.

Training sample number	3	4	5	6
LBP	86.93	91.08	94.10	96.00
CS-LBP	88.09	91.70	94.40	96.00
Gabor + LBP	89.78	93.32	95.54	96.89
MFGC + CS-LBP	88.36	91.79	93.50	94.69
MOGC + CS-LBP	90.32	93.54	95.65	97.00

Referring to Table 3, (1) CS-LBP algorithm showed higher recognition rate than LBP algorithm in other training samples, (2) MFGC + CS-LBP algorithm, CS-LBP algorithm showed similar recognition rate, ( 3) MOGC + CS-LBP algorithm achieved the best recognition speed.

Next, look at the test results of the FERET face database. 120 people were selected from the FERET facial standard library, six images each and a total of 720 facial images. Each image you select includes a change of expression, lighting, and age. Referring to FIG. 8, one of six images is selected from the FERET face standard library. That is, Figure 8 shows six sample images of the person of FERET.

In the test, two and four images each were randomly selected as training samples and the rest were selected as test samples. Table 4 shows the test comparison results for five different algorithms according to the iterative test and the 10-fold average test. That is, Table 4 shows the recognition rate results of the five algorithms in FERET.

Training sample number	2	3	4
LBP	82.79	88.86	92.33
CS-LBP	82.21	89.00	91.79
Gabor + LBP	86.70	91.33	92.89
MFGC + CS-LBP	85.86	90.03	92.00
MOGC + CS-LBP	87.37	91.42	93.00

Referring to Table 4, (1) LBP and CS-LBP algorithms achieve similar recognition rates, and (2) Gabor + LBP and MOGC + CS-LBP algorithms still have significant recognition rates.

As a result of the analysis of the test results, the MOGC + CS-LBP algorithm has a low level of feature extraction and a significant recognition rate compared to the Gabor + LBP algorithm. That is, the MOGC + CS-LBP algorithm has the best recognition rate and the accuracy of the recognition rate is improved.

The invention can also be embodied as computer readable code on a computer readable recording medium. Computer-readable recording media include all kinds of recording devices that store data that can be read by a computer system.

Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disks, optical data storage devices, and the like, which are also implemented in the form of carrier waves (eg, transmission over the Internet). It also includes.

The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion. And functional programs, codes and code segments for implementing the present invention can be easily inferred by programmers in the art to which the present invention belongs.

As described above, the present specification and drawings have been described with respect to preferred embodiments of the present invention, although specific terms are used, it is only used in a general sense to easily explain the technical contents of the present invention and to help the understanding of the present invention. It is not intended to limit the scope of the present invention. It will be apparent to those skilled in the art that other modifications based on the technical idea of the present invention can be carried out in addition to the embodiments disclosed herein.

The present invention provides an improved multi-channel Gabor for human face recognition combining Gabor filter and Center-symmetric local binary patterns (CS-LBP) to reduce the noise's robust feature point extraction and the high dimensionality of the extracted facial feature points. The filter-based human face recognition method,

By combining the Gabor filter with CS-LBP (Center-symmetric local binary patterns), it provides noise-resistant feature point extraction and high dimensionality of extracted facial feature points.

In addition, the improved multi-channel Gabor filter-based human face recognition method according to another embodiment of the present invention reduces the dimension of the feature image by combining the Gabor feature images in different directions and scales, and based on CS-LBP based low dimension from the feature image. Provides the effect of extracting facial feature points.

In addition, the improved multi-channel Gabor filter-based human face recognition method according to another embodiment of the present invention is characterized by a feature dimension, a storage space, and a combination of the Gabor filter and the CS-LBP method, compared to the conventional Gabor filter and LBP approach. It reduces the computation time and at the same time provides the effect of high accuracy of face recognition.

Claims (4)

1. Face image of N inputs (N is a natural number of 2 or more)

   

   Output is an image of a cascading particle histogram feature vector representation of the statistics.

   

   Feature image

   

   (

   

   ,

   

   A first step of obtaining;

   bracket

   

   In the same manner, a second step of separating a plurality of overlapping sub-images; And

   

   Generating a histogram of all sub-images of the, and extracting a cascading histogram feature vector sequence; An improved multi-channel Gabor filter-based human face recognition method comprising a.
2. The method according to claim 1, After the third step,

   each

   

   Extracting a corresponding eigenvector of the cascade sequence of; An improved multi-channel Gabor filter based human face recognition method further comprising.
3. The method according to claim 2, After the fourth step,

   A fifth step of extracting a face image feature vector after the feature extraction of the first to fourth steps; An improved multi-channel Gabor filter-based human face recognition method further comprising.
4. The method according to claim 1, wherein the first step,

   Face image

   

   Step 1-1 of obtaining a heirloom amplitude spectrum by performing heirloom conversion;

   For the amplitude spectrum of n different by overlap,

   

   Obtaining a first step 1-2; And

   Image for each CS-LBP code

   

   Obtaining the first 1-3 steps; An improved multi-channel Gabor filter-based human face recognition method comprising a.

Images