KR20150069989A - Apparatus for simultaneously detecting and recognizing a face using local block texture feature - Google Patents

Abstract

The present invention relates to a method for detecting and recognizing a face for an automated recognition system such as artificial intelligence and an intelligent surveillance system. The method includes the steps of: extracting a local block texture feature map from an input image; detecting a face area from the extracted local block texture feature map; and recognizing the face by using the detected face. According to an embodiment of the present invention, when a face is detected and recognized, detection and recognition of a face are executed at the same time by using an integrated local block texture feature vector, extracted from the input image. Accordingly, detection and recognition of a face can be more efficiently executed even in an environment with a little resource.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a face detection and recognition apparatus using a local block texture feature vector,

The present invention relates to face recognition. More particularly, in detecting and recognizing a face from an input image, face detection and face recognition are simultaneously performed using a local block texture feature vector extracted from an input image, And more particularly, to a face detection and recognition apparatus capable of effectively performing face detection and recognition even in a resource environment.

Recently, face recognition technology has attracted attention as an element technology in artificial intelligence, privacy protection and intelligent surveillance system application. As a result, many researches on element technology for face recognition have been conducted. Basically, the face recognition technology detects the face region with respect to the input image, and recognizes the face by comparing the detected face region with the training set.

The research on the conventional face recognition is divided into the research on the face detection and the study on the face recognition, and is performed by different feature vectors for detection and recognition. However, in a system that requires simultaneous face detection and recognition, detection and recognition are performed by different feature vectors is not efficient considering the final system speed and the amount of calculation required for the system.

In order to reduce the system speed and calculation amount, there is an increasing demand for light weight of various detection and recognition algorithms, and a facial feature or recognition system is required to obtain high performance for face detection and recognition in a real environment. However, there still exists a trade-off between efficiency in terms of calculation and cost for enhancing face detection and recognition performance. Accordingly, it is a task to develop an optimal face detection and face recognition system between trade-offs.

(Patent Literature)

Korean Registered Patent No. 10-0795160 (Registered Date Jan. 09, 2008)

Therefore, in detecting and recognizing a face from an input image, the present invention simultaneously performs face detection and face recognition using a local block texture feature vector extracted from an input image, thereby efficiently performing face detection and recognition in a small resource environment And a face detection and recognition apparatus capable of performing face detection and recognition.

The present invention relates to a face detection and recognition apparatus, comprising: an image input unit for receiving an input image input from a camera; a texture feature map extracting unit for extracting a texture feature map from an image input through the image input unit; A face detection unit for detecting a face area in the input image based on the texture feature map; and a face recognition unit for recognizing a face in the face area based on the detected face area.

The texture feature map extraction unit may divide the input image into a plurality of local regions, generate a binary code by comparing magnitudes between pixel values centered and neighboring to each local region, And calculating a texture feature value through the evolution, and calculating a texture feature map expressed by the texture feature value with respect to the input image.

The face detection unit may calculate a local block texture feature vector for a window having a predetermined size scanned for face detection in the input image based on the texture feature map, And a face region is detected in the input image.

The face detection unit may generate a binary map from the texture feature map, extract an integral feature map by integrating the binary map, extract a local block texture feature vector for all local blocks of the input image from the integral feature map, Is calculated.

The face detection unit may check whether there is a value corresponding to each histogram bin based on the texture feature map and determine whether the texture feature map has a value corresponding to each histogram bin based on the inspection result As the binary map.

The face detector may scan a window of a predetermined size for each input image based on the integration feature map, count the frequency of the local block texture feature value for n local block predefined within each window, A histogram vector corresponding to a local block is generated, and a histogram vector for each local block is concatenated to calculate a local block texture feature vector corresponding to the window.

The face detection unit may calculate a Euclidean distance between the local block texture feature vector for the window and the reference feature vector, and compare the Euclidean distance with a preset threshold value. Then, based on the comparison result, Is detected.

The face recognition unit may calculate the Euclidean distance between the local block texture feature vector and each template feature vector through matching between the local block texture feature vector and the face database stored in the template feature vector of the recognition target faces, Comparing the threshold value with the set threshold value, and recognizing the face in the face area based on the comparison result.

According to another aspect of the present invention, there is provided a face detection and recognition method comprising: receiving an input image input from a camera; extracting a texture feature map from the input image; Detecting a face region in the face region based on the detected face region;

The extracting of the texture feature map may further include dividing the input image into a plurality of local regions, generating a binary code by comparing magnitudes between pixel values adjacent to the center of each local region, Calculating a texture feature value through decimation of the generated binary code, and calculating a texture feature map expressed by the texture feature value with respect to the input image.

The step of detecting the face region may include calculating a local block texture feature vector for a window having a predetermined size scanned for face detection in the input image based on the texture feature map, And detecting a face region in the input image by matching a vector with a predetermined reference feature vector.

The step of calculating the local block texture feature vector may further include generating a binary map from the texture feature map, integrating the binary map to extract an integration feature map, And calculating a local block texture feature vector for all local blocks of the block.

The generating of the binary map may include: checking whether there is a value corresponding to each histogram bin based on the texture feature map; and determining whether the texture feature map is included in each of the histogram bins And generating the binary map indicating whether there is a corresponding value.

The calculating of the local block texture feature vector may include scanning a window of a predetermined size for each input image based on the integration feature map, Generating a histogram vector corresponding to each local block by counting the frequency of the local block texture feature value; and calculating a local block texture feature vector corresponding to the window by concatenating the histogram vectors for each local block .

The recognizing of the face is a step of recognizing a face in the detected face region by performing matching between the local block texture feature vector and the face database.

The step of recognizing the face may further include performing matching between the local block texture feature vector and the stored face database of template feature vectors of recognized faces, Calculating a Euclidean distance between the template feature vectors of the Euclidean distance and a face value of the face region by comparing the calculated Euclidean distance with a preset threshold value.

According to the embodiment of the present invention, in detecting and recognizing a face, face detection and face recognition are simultaneously performed using an integrated local block texture feature vector extracted from an input image, thereby enabling face detection and recognition There is an advantage that it can be performed more efficiently.

1 is a block diagram of a system for simultaneously performing face detection and recognition using a local block texture feature map according to an embodiment of the present invention;
FIG. 2 is an internal configuration diagram of a texture feature map extracting unit according to an embodiment of the present invention;
3 is a flowchart of an operation control for detecting a face from a texture feature map according to an embodiment of the present invention;
FIG. 4 is a flowchart of an operation control for recognizing a face from a texture feature map according to an embodiment of the present invention;
Figure 5 illustrates an example of a localized binary pattern according to an embodiment of the present invention,
6 is a diagram illustrating an example of generating a histogram vector using texture feature values according to an embodiment of the present invention.
FIG. 7 illustrates a process of deriving an integral feature map from an input image according to an embodiment of the present invention; FIG.
8 is an exemplary diagram illustrating the relationship of R, G, and B color pixel values that determine skin color according to an embodiment of the present invention;
9 is a conceptual diagram illustrating an operation of generating a texture feature map having a size smaller than an input size using feature values of a texture feature map according to an exemplary embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention and the manner of achieving them will become apparent with reference to the embodiments described in detail below with reference to the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. To fully disclose the scope of the invention to those skilled in the art, and the invention is only defined by the scope of the claims.

In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. The following terms are defined in consideration of the functions in the embodiments of the present invention, which may vary depending on the intention of the user, the intention or the custom of the operator. Therefore, the definition should be based on the contents throughout this specification.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

FIG. 1 is a block diagram of an apparatus for simultaneously performing face detection and recognition using a local block texture feature map according to an embodiment of the present invention. The image input unit 100, the texture feature map extraction unit 200, (300), a face recognition unit (400), and the like.

Hereinafter, the operation of each component of the facial recognition apparatus of the present invention will be described in detail with reference to FIG.

First, the image input unit 100 receives an image including a face to be detected and recognized by the camera. The video input unit 100 includes a CCTV or a black box video input to an on-line video or a content stored in an arbitrary remote or local storage device. The content file type of JPEG, BMP, GIF, etc., which are currently widely used as the file type of the media content of the present invention, is basically included, but is not limited to the content file type of the above example.

The texture feature map extracting unit 200 extracts a local texture feature map from the image input through the image input unit 100.

FIG. 2 shows the internal structure of the texture feature map extracting unit 200. 2, in the texture feature map extracting unit 200, in order to extract a texture feature of an image input through the camera image input unit 100, based on a local block texture feature pattern, The texture feature value 202 is extracted.

At this time, an example of a typical local block texture pattern is a local binary pattern. In order to generate the texture feature map using the local binary pattern values, as shown in FIG. 5, a binary code is obtained by comparing magnitudes between center values and neighboring pixel values for each local region in the input image, The feature value is calculated through decimation of the sign. At this time, the texture feature map means an image having a texture feature value obtained from the input image. After the calculation of the texture feature value, an image having a texture feature value is obtained with respect to pixel values of all the input images. The texture feature map including the texture feature values is input to the face detection unit 300 and the face recognition unit 400.

The face detecting unit 300 detects a face region in the input image based on the texture feature map. That is, the face detection unit 300 calculates a local block texture feature vector for a window of a predetermined size scanned for face detection on the input image based on the texture feature map, and sets a local block texture feature vector And the face region is detected from the input image by matching with the reference feature vector.

In addition, the face detection unit 300 generates a binary map from the texture feature map, integrates the generated binary map to extract an integral feature map, extracts an integral feature map from the extracted integral feature map, The block texture feature vector is calculated.

At this time, the face detector 300 scans a window of a predetermined size for each input image based on the integration feature map, counts the frequency of local block texture feature values for n local blocks defined in each window, A histogram vector corresponding to a local block is generated, and a histogram vector for each local block is concatenated to calculate a local block texture feature vector corresponding to the window.

3 illustrates an operation control flow for detecting a face in an input image using a texture feature map in a face detection unit according to an embodiment of the present invention.

Hereinafter, the operation of the face detecting unit 300 will be described in more detail with reference to FIG.

First, the face detecting unit 300 receives a texture feature map from the texture feature map extracting unit 200 (S3000), and creates a texture feature vector for all the scanned blocks for face detection based on the texture feature map ( S3100). In order to effectively detect the face region and fully utilize the structural information of the face and to make the pose change robust, j predefined j number of overlapped local blocks (L ₁ , L ₂ , ... L _j ), the histogram vector can be generated by calculating the frequency of the texture feature values. In Fig. 6, the block j = 3 is expressed.

)에 대해 얻은 질감 특징 맵(

)을 입력받고, 이러한 질감 특징 맵에 기반하여 각 히스토그램 빈(bin, n)에 해당하는 값이 있는지 여부를 [수학식 1]과 같이 나타내는 이진 맵(binary map,

)을 얻게 된다. 이때,

함수는

값이 n 과 같을 때 1을 반환하는 이진 함수로 정의된다.In order to efficiently calculate this feature vector,

) ≪ / RTI >

), And determines whether or not there is a value corresponding to each of the histogram bin (bin, n ) based on the texture feature map as a binary map (binary map)

). At this time,

The function

It is defined as a binary function that returns 1 when the value is equal to n .

)을 위의 [수학식 2]에 의해 얻게 되어 적분 특징 맵을 추출한다. 이어, 적분 특징 맵을 기반으로 일정 크기의 윈도우를 스캔하면서 각 윈도우 내에서 미리 정의된 j개의 블록(

)에 대해 국부 블록 질감 특징 값의 빈도를 각 히스토그램 빈에 대해 단순히 2번의 덧셈과 뺄셈 연산을 이용하여 각 블록에 해당하는 히스토그램 벡터(

)를 얻게 된다. 만약 i번째 블록

을 이루는 좌표가

그리고

라면, 히스토그램 벡터는 다음 [수학식 3]에 의해 얻어지게 된다. 이 때의 연산은 벡터의 전치(transpose)를 의미한다. 히스토그램 벡터에서의 각 성분은

에 의해 얻어진다.For this binary map, the integral image (

) Is obtained by the above expression (2), and the integral feature map is extracted. Next, a predetermined size window is scanned based on the integral feature map, and j blocks (

), The frequency of the local block texture feature value is added to the histogram vector corresponding to each block by simply adding and subtracting two times to each histogram bin

). If the i- th block

The coordinates that make up

And

, The histogram vector is obtained by the following equation (3). The operation at this time is the transpose of the vector. Each component in the histogram vector

Lt; / RTI >

)는 [수학식 4]와 같이 계산된다. Finally, by connecting the histogram vectors for each block, the texture feature vector corresponding to the window to be scanned

) Is calculated as shown in Equation (4).

) 사이의 유클리드 거리(

)를 계산하고, 계산된 유클리드 거리를 미리 정의된 임계값과 비교하여 얼굴 영역인지 아닌지 판단하게 된다.Then, the face detecting unit 300 detects the face region by matching based on the texture feature vector (S3300). In this matching, as the simplest method, for example, template matching is performed using a texture feature vector and a feature vector of a template defined in advance

) Between the Euclidean distance

), And compares the calculated Euclidean distance with a predefined threshold to determine whether or not it is a face region.

연산은 벡터의

을 의미한다. In Equation (5) above,

The operation

.

Then, the face detecting unit 300 removes candidate regions judged as false positives (FP) among the finally detected candidate regions (S340), and outputs the final face detection result (S3500). At this time, in order to set a criterion for judging whether or not the candidate region is erroneously detected, for example, a relation of R, G, B color pixel values which determine the skin color defined in FIG. 8 can be used.

At this time, the recognizer for determining the face region can be flexibly replaced by the template matching or AdaBoost (adaptive boosting) recognizer.

Referring to FIG. 3, it is possible to detect various sizes of facial regions by scanning an image having various sizes using an input image. For each size, Extraction is not computationally efficient.

Accordingly, the face detecting unit 300 can use a feature map scaling method that generates a texture feature map having a size smaller than the input size by using only the feature values of the texture feature map of the input size. In this method, as shown in FIG. 9, the image coordinates of the small-sized feature map to be estimated are used as feature values around the point projected into the texture feature map of the input size. At this time, the neighboring feature values can be represented by a binary code. Feature values of feature maps corresponding to a small size are estimated by a majority vote of 1 or 0 for each bit of the binary code. Therefore, it is not necessary to extract a texture feature map for various sizes by generating an estimated texture feature map for various sizes, so that efficient computation is possible.

FIG. 4 illustrates an operation control flow for recognizing a face by reusing a texture feature vector used in face detection in the face recognition unit 400 according to an embodiment of the present invention.

That is, the face recognition unit 400 receives the texture feature map from the texture feature map extraction unit 200 for the input image to be subjected to the face recognition (S4000) The feature vector is received (S4100).

Next, the face recognition unit 400 performs face recognition by reusing the texture feature map and the texture feature vector calculated for the face area detection in the face detection unit 300 (S4200).

That is, the face recognition unit 400 performs face recognition through matching between a feature vector and a previously prepared data base, and outputs a face recognition result corresponding to the closest identity information according to the matching result (S4300). At this time, for example, the template feature vector of the target faces to be recognized may be stored on the face database. In the face detection, the Euclidean distance between the texture feature vector and the template feature vector is compared with a predetermined threshold value, The face can be recognized in the area.

The face recognizing unit 400 can be flexibly adopted as another recognizing unit to enhance the face recognizing performance as in the face detecting unit 400 described in FIG.

As described above, according to the embodiment of the present invention, the face detection and recognition can be efficiently performed simultaneously by using the integrated local block texture feature vector for face detection and recognition.

The foregoing description is merely illustrative of the technical idea of the present invention and various changes and modifications may be made by those skilled in the art without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the scope of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

100: image input unit 200: texture feature map extracting unit
300: face detecting unit 400: face recognizing unit

Claims (3)

An image input unit for receiving an input image input from a camera,
A texture feature map extracting unit for extracting a texture feature map from an image input through the image input unit,
A face detection unit for detecting a face region in the input image based on the extracted texture feature map;
A face recognition unit for recognizing a face in the face area based on the detected face area,
And a face detection and recognition device.
The method according to claim 1,
The texture feature map extracting unit extracts,
The input image is divided into a plurality of local regions, a binary code is generated by comparing magnitudes between pixel values centered and neighbor to each local region, a texture feature value is calculated through decimation of the generated binary code And calculates a texture feature map expressed by the texture feature value with respect to the input image.
The method according to claim 1,
Wherein the face detection unit comprises:
Calculating a local block texture feature vector for a window having a predetermined size scanned for face detection in the input image based on the texture feature map, matching the local block texture feature vector with a preset reference feature vector, And the face detection unit detects the face area in the face detection unit.

Images