CN106022223B - A high-dimensional partial binary pattern face recognition method and system - Google Patents

Abstract

The invention discloses a high-dimensional local binary mode face recognition algorithm and system, the algorithm comprising the following steps: S1, acquiring a face image, and preprocessing it to obtain a grayscale image of the same size; The processed grayscale image is subjected to HDLBP feature extraction to obtain the corresponding feature image; S3. Extract the histogram of the feature image to obtain the corresponding feature vector; S4. Compare the feature vector with the information in the feature database to obtain the recognition result. The invention can extract the local features and the global features of the image, greatly improving the recognition rate of the algorithm; and increasing the accuracy rate of the image recognition on the basis of ensuring that the complexity of the algorithm is not high.

Description

A high-dimensional partial binary pattern face recognition method and system

technical field

The invention relates to the field of face recognition, in particular to a face recognition method and system of a high-dimensional local binary pattern.

Background technique

Local Binary pattern (LBP) algorithm, as a face recognition algorithm, is an algorithm that relies on local texture description proposed by Ojala and Ahonen et al. in 1996, and is used to describe pixels and pixels in their neighborhood. The numerical relationship between points is widely used in the field of face recognition because of its concise calculation method, good description of local features of the image, and insensitivity to illumination. At the same time, because the LBP descriptor only pays attention to the description of the local features of the image, and ignores the description of the global features of the image, it leads to the shortage of the global feature extraction of the LBP algorithm. In order to effectively solve this problem, many scholars have studied it and proposed many improvement and optimization methods.

Zhou Xi et al. proposed a method of block processing in order to solve the problem that LBP descriptors are insufficient in extracting global features. The core idea of split processing is to divide the original image into equal parts according to a certain size, or divide the original image according to the location of human facial features, extract LBP features from the sub-images, concatenate all the feature vectors together, and obtain the global According to some characteristics, the performance of the LBP algorithm after sub-processing is better than that of the unprocessed LBP algorithm through experiments, but there is no definite answer on the point of dividing the original image according to what standard. Neither can get a good effect on the processing of the original image, so Wang Hong et al. conduct research from the LBP descriptor itself, the idea is to enlarge the LBP descriptor by equal times, so that it can extract images in a wider range feature. The comparison is no longer a certain pixel, but the average value of the pixels in the fixed neighborhood including a certain pixel is calculated, which ensures that the local characteristics also reflect the global characteristics to a certain extent, but for the neighbor The choice of domain size has become the focus of research again. Also starting from the BLP descriptor itself, different from the principle of equal magnification, Wang Cheng et al. reflect the global features through the features in the multi-level neighborhood of the multi-scale weighted center point. The closer the point in the neighborhood is to the center point, the weighted The heavier the weight, the more weighted levels, the better the globality, and the higher the computational complexity.

Contents of the invention

The technical problem to be solved by the present invention is to provide a high-dimensional local binary mode face recognition method and system that can greatly improve the recognition rate of the algorithm for the defect that the global feature cannot be extracted in the prior art.

The technical solution adopted by the present invention to solve its technical problems is:

The present invention provides a kind of face recognition algorithm of high dimension partial binary value mode, comprises the following steps:

S1. Obtain a face image, and preprocess it to obtain a grayscale image of the same size;

S2. Perform HDLBP feature extraction on the preprocessed grayscale image to obtain a corresponding feature image;

S3. Extracting the histogram of the feature image to obtain a corresponding feature vector;

S4. Comparing the feature vector with information in the feature database to obtain a recognition result.

Further, the method for preprocessing the grayscale image in step S1 of the present invention is specifically:

Let the distribution of the local texture V of the face grayscale image be:

V=v(g _c g ₀ ... g _p-1 g)

Among them, g _c represents the central threshold of the window, k=0,2...p-1 in g _k , g _k represents the gray value of each neighborhood pixel, p represents the number of neighborhood points, and g represents the face The gray mean value of the grayscale image, the calculation formula is:

Among them, m×n is the size of the grayscale image, and g(i j) is the grayscale value of each pixel in the image.

Further, the method for HDLBP feature extraction in step S2 of the present invention is specifically:

When calculating the HDLBP descriptor, firstly, the calculation method of the classic LBP descriptor in the window is used to ensure the local features; The calculation method guarantees the global features; finally, the local features are integrated as low-dimensional and global features as high-dimensional, and the calculation result is the feature value of the central pixel in the window.

Further, the method for performing high-dimensional and low-dimensional fusion in step S2 of the present invention is specifically:

The central feature is added to the binary sequence of the edge feature as the highest dimension component, so that the feature sequence is extended to a higher dimension, and the amount of information contained in the feature sequence is expanded; the low-dimensional features and high-dimensional features are fused together according to the following formula, so that Two columns of feature sequences become one column of feature sequences, and the corresponding feature values can be obtained by calculating according to the method of converting binary to decimal; the calculation formula is:

Among them, g _c represents the central threshold of the window, k=0,2...p-1 in g _k , g _k represents the gray value of each neighborhood pixel, p represents the number of neighborhood points, and g represents the face The gray mean value of the grayscale image, the calculation formula is:

Among them, m×n is the size of the grayscale image, and g(i j) is the grayscale value of each pixel in the image.

Further, the formula for HDLBP feature extraction in step S2 of the present invention is:

Among them, the s function is as follows:

Further, the method for extracting the histogram of the feature image in step S3 of the present invention is specifically:

According to the input feature image, sort all the pixels in the image in ascending order according to their gray value, and then count the number of occurrences of pixels with the same gray value to obtain an n*1 sequence, which is the feature vector, where n represents the number of different gray values in the feature image, and the calculation formula of the histogram is as follows:

h(i)=NUM(g _i )i∈(1,n)

Among them, h(i) represents the number of pixels whose gray value is g _i .

Further, the method for obtaining the recognition result in step S4 of the present invention is specifically:

According to the comparison between the feature vector and the information in the feature database, the Euclidean distance is used as the measure, and the nearest neighbor classification method is used for identification.

The invention provides a high-dimensional local binary pattern face recognition system, comprising:

An image preprocessing unit is used to obtain a face image and preprocess it to obtain a grayscale image of the same size;

The HDLBP feature extraction unit is used to extract the HDLBP feature from the preprocessed grayscale image to obtain a corresponding feature image;

A feature vector extraction unit is used to extract the histogram of the feature image to obtain a corresponding feature vector;

The image recognition unit is used to compare the feature vector with information in the feature database to obtain a recognition result.

The beneficial effects produced by the present invention are: the high-dimensional local binary mode face recognition algorithm of the present invention ensures the local features by extending the calculation method of the classic LBP descriptor in the window; The same calculation method is used to ensure the global features; finally, the local features are regarded as low-dimensional, and the global features are integrated as high-dimensional, as the feature value of the central pixel in the window; The invented algorithm can extract the local and global features of the image, greatly improving the recognition rate of the algorithm; and on the basis of ensuring that the complexity of the algorithm is not high, the accuracy of image recognition is increased.

Description of drawings

The present invention will be further described below in conjunction with accompanying drawing and embodiment, in the accompanying drawing:

Fig. 1 is the flow chart of the face recognition algorithm of the high-dimensional local binary pattern of the embodiment of the present invention;

Fig. 2 is the fusion process of the high-dimensional local binary pattern face recognition algorithm of the embodiment of the present invention;

Fig. 3 is the block diagram of the face recognition algorithm of the high-dimensional local binary pattern of the embodiment of the present invention;

Fig. 4 is the calculation process diagram of the high-dimensional local binary pattern face recognition algorithm of the embodiment of the present invention;

FIG. 5 is a block diagram of a high-dimensional local binary pattern face recognition system according to an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

As shown in Figure 1, the high-dimensional local binary pattern face recognition algorithm of the embodiment of the present invention comprises the following steps:

S1. Obtain a face image, and preprocess it to obtain a grayscale image of the same size; the method of preprocessing to obtain a grayscale image is specifically:

Let the distribution of the local texture V of the face grayscale image be:

V=v(g _c g ₀ ... g _p-1 g)

Among them, g _c represents the central threshold of the window, k=0,2...p-1 in g _k , g _k represents the gray value of each neighborhood pixel, p represents the number of neighborhood points, and g represents the face The gray mean value of the grayscale image, the calculation formula is:

Among them, m×n is the size of the grayscale image, and g(i j) is the grayscale value of each pixel in the image.

S2. Carry out HDLBP feature extraction to the preprocessed grayscale image to obtain a corresponding feature image; the formula for HDLBP feature extraction is:

Among them, the s function is as follows:

S3. Extracting the histogram of the feature image to obtain a corresponding feature vector;

S4. Comparing the feature vector with information in the feature database to obtain a recognition result.

When calculating the HDLBP descriptor, firstly, the calculation method of the classic LBP descriptor in the window is used to ensure the local features; The calculation method guarantees the global features; finally, the local features are integrated as low-dimensional and global features as high-dimensional, and the calculation result is the feature value of the central pixel in the window.

As shown in Figure 2, the fusion process is as follows:

The central feature is added to the binary sequence of the edge features as the highest dimension component, so that the feature sequence is extended to a higher dimension, and the feature sequence contains a larger amount of information. Since the low-dimensional features and high-dimensional features are fused together, the two columns of feature sequences become one column of feature sequences. Therefore, when converting the feature sequence into a feature value, it only needs to be calculated according to the binary to decimal method to obtain the corresponding feature value. , the calculation formula is:

As shown in Figure 3, in another specific embodiment of the present invention,

The specific calculation process of the algorithm is as follows:

Step 1: Input the face image to be recognized;

Step 2: Preprocess the input face image to obtain a grayscale image of the same size;

The third step: extract the HDLBP feature of the face image to be recognized according to the following formula, and obtain the corresponding feature image;

According to the input grayscale image, calculate the value of g according to the formula for calculating g as follows, and then use the calculation formula of HDLBP feature to scan the grayscale image from left to right and top to bottom, for each i, j Each value has a corresponding eigenvalue, and the image composed of eigenvalues is the feature image.

Step 4: Extract the histogram of the feature image to obtain the corresponding feature vector;

According to the input feature image, sort all the pixels in the image in ascending order according to their gray value, and then count the number of occurrences of pixels with the same gray value to obtain an n*1 sequence, which is the feature vector, where n represents the number of different gray values in the feature image, and the calculation formula of the histogram is as follows:

h(i)=NUM(g _i )i∈(1,n)

Among them, h(i) represents the number of pixels whose gray value is g _i .

Step 5: On the established feature database, use the Euclidean distance as a measure, and use the nearest neighbor classification method to identify;

The Euclidean distance calculation formula in n-dimensional space is as follows:

Where x _1k and x _2k are the kth components of n-dimensional vectors x ₁ and x ₂ respectively. Nearest neighbor classification method: divide the data to be classified into the class of the nearest sample according to the distance.

Use the Euclidean distance formula to calculate the Euclidean distance between the feature vector and all samples in the feature database, and the sample with the smallest distance is the result of recognition. Among them, d ^m is the Euclidean distance between the feature vector x _k and the mth sample y _k in the database.

Step 6: Output the recognition result.

As shown in Figure 4, the ORL face database and the YALE face database were used for experiments. The ORL face database was created by the Media AT&T Laboratory of the University of Cambridge. The library includes 40 categories of faces, each category has 10 grayscale images with a size of 112×92. All images have similar dark backgrounds, and different images of the same person are taken at different times, under different lighting, different head postures, different expressions, and different details; the YALE face database is a database established by Yale University in the United States. Face database, the database contains 15 people, each with 11 images, including different expressions, different lighting directions, and changes in details. The face database is uniformly divided into test samples and experimental samples. For ORL face database, test sample:experimental sample=5:5; for YALE face database, test sample:experimental sample=6:5.

The experimental results are shown in the table below.

It can be seen from the above that whether it is in the YALE face database or the ORL face database, the recognition rate of the high-dimensional local binary pattern algorithm is higher than that of the traditional local binary pattern.

Aiming at the shortcoming that the traditional local binary pattern face recognition algorithm cannot extract global features, a face recognition algorithm based on high-dimensional local binary pattern is proposed. Compared with the traditional local binary pattern algorithm, this algorithm greatly improves the recognition rate of the algorithm.

As shown in Figure 5, the high-dimensional local binary pattern face recognition system of the embodiment of the present invention is used to realize the high-dimensional local binary pattern face recognition algorithm of the embodiment of the present invention, including:

An image preprocessing unit is used to obtain a face image and preprocess it to obtain a grayscale image of the same size;

The HDLBP feature extraction unit is used to extract the HDLBP feature from the preprocessed grayscale image to obtain a corresponding feature image;

A feature vector extraction unit is used to extract the histogram of the feature image to obtain a corresponding feature vector;

The image recognition unit is used to compare the feature vector with information in the feature database to obtain a recognition result.

It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should belong to the protection scope of the appended claims of the present invention.

Claims (3)

1. a kind of higher-dimension local binary patterns face identification method, which comprises the following steps:

S1, facial image is obtained, and it is pre-processed to obtain the gray level image of identical size；

S2, HDLBP feature extraction is carried out to pretreated gray level image, obtains corresponding characteristic image；

S3, the histogram for extracting characteristic image, obtain corresponding feature vector；

S4, it is compared according to feature vector with the information in property data base, obtains recognition result；

The method of HDLBP feature extraction is carried out in step S2 specifically:

HDLBP description has adopted calculation method of LBP description in window first, ensure that local feature when calculating； Then calculating of the son in window is described with LBP to the gray average of window center pixel gray value and face gray level image Method guarantees global characteristics；Finally using local feature as low-dimensional, global characteristics are combined as higher-dimension, the knot of calculating Fruit is exactly the characteristic value of central pixel point in the window；

The method that higher-dimension and low-dimensional fusion are carried out in step S2 specifically:

It is added to central feature as the component of most higher-dimension in the binary sequence of edge feature, so that characteristic sequence Xiang Gaoyi Dimension stretching, extension expands the information content that characteristic sequence includes；Low-dimensional feature and high dimensional feature are fused together according to the following formula, made Two column characteristic sequences become a column characteristic sequence, and according to binary system turn metric method calculate can be obtained it is corresponding Characteristic value；Calculation formula are as follows:

Wherein, g_cRepresent the central threshold of window, g_kMiddle k=0,2...p-1, g_kIndicate the gray value of each neighborhood territory pixel point, p table Show neighborhood point number, g indicates the gray average of face gray level image, calculation formula are as follows:

Wherein, m × n is the size of gray level image, and g (i j) is the gray value of each pixel in image；

The formula of HDLBP feature extraction is carried out in step S2 are as follows:

Wherein, s function is as follows:

2. higher-dimension local binary patterns face identification method according to claim 1, which is characterized in that extracted in step S3 The method of the histogram of characteristic image specifically:

According to the characteristic image of input, pixel all in image is subjected to ascending sort according to the size of its gray value, so The number that there is statistics the pixel of same grayscale value to occur afterwards, obtains a n*1 sequence, i.e. feature vector, and wherein n indicates special The number of different gray values in image is levied, the calculation formula of histogram is as follows:

H (i)=NUM (g_i)i∈(1,n)

Wherein h (i) indicates that gray value is g_iPixel number.

3. higher-dimension local binary patterns face identification method according to claim 1, which is characterized in that obtained in step S4 The method of recognition result specifically:

It is compared according to feature vector with the information in property data base, is to measure with Euclidean distance, utilizes nearest neighbour classification Method is identified.