RU2541918C2 - Apparatus for automated facial recognition when processing group photograph - Google Patents

Abstract

FIELD: physics, computer engineering.

SUBSTANCE: invention relates to computer engineering and can be used in video analysis and processing systems. The apparatus for automated facial recognition when processing a group photograph comprises a unit for searching for regions containing faces, a decision unit, wherein the output of a unit for localising facial regions (1) is connected to the input of a unit for separating the image into a background and an object (2), the output of which is connected to the first input of an image normalisation unit (3); the output of a control unit (9) is connected to the second input of the image normalisation unit (3), the output of which is connected to the input of a unit for generating vectors of facial features (4), the output of which is connected to the input of a buffer unit (5), the output of which is connected to the first input of a comparator unit (6); the output of a delay unit (7) is connected to the input of a unit for generating vectors of facial features (8), the output of which is connected to the second input of the comparator unit (6), the output of which is the data output of the apparatus.

EFFECT: automated facial recognition when processing a group photograph.

1 dwg

Description

The invention relates to the field of computer technology and can be used in digital television and photo systems of global positioning and surveillance systems.

The mathematical model of the observed image I is determined by the following expression:

I = α ₁ I ₁ + (1-α ₁ ) · (α ₂ I ₂ + (1-α ₂ ) (... + (1-α _N-1 ) · I _N )) (1)

where I ₁ , ... I _N is the set of image layers separated by alpha channels α _N ∈ [0,1], the dimensions I ₁ , ..., I _N and α _N coincide. Alpha channels α _N are masks that allow you to select individual objects in the scene. The solution to the problem of face recognition in group digital photography is to establish the correspondence of each I _N image to a given image of a person's face.

The main task to be solved is recognition of a person’s face in a group photo.

Of considerable interest in the field of computer vision is the task of detecting and recognizing faces when processing multidimensional signals (images and video sequences). The scope of the solution of this problem is crime prevention systems and identification of criminals, verification of people at airports, automatic security systems with “face control”, personalization of home appliances, forensic examination of photo and video filming, virtual reality, computer games, etc.

Existing recognition methods are divided into two classes: methods based on the search and comparison of local features of the face (forming a vector of descriptors based on local features) and decomposition of the local area of the face image into orthogonal components (EigenFace method based on obtaining the main components). As a rule, the use of the methods considered is applicable to multidimensional signals (images) containing only one face on a uniform background. The use of these methods for processing group images of people is extremely limited. In this regard, it is relevant to solve the problem of recognizing a group of faces in images with a textured background.

A known method of identifying a person’s identity by digital face image [Patent No. 2431191, IPC G06K 9/00]. The invention relates to automation and computer technology and can be used in artificial intelligence systems: access control systems, robotic systems that interact with a human operator in a production environment and other systems. The technical result is to increase the accuracy and efficiency of facial image recognition. The method of identifying a person’s identity by digital image of a person includes the steps of searching for a person’s face in the frame, scaling the selected image to a given size, aligning the brightness and color of the image, forming the vector of the input image (P1), downloading from the image database for comparison (P2), splitting the image on the blocks and the implementation of the primary search for the coincidence of these blocks, the implementation of the secondary search, on the basis of which a decision is made on compliance, and after loading the image oditsya alignment of luma and chroma, then the vector of the input image is formed, and then the image is scaled to a predetermined size and then divide the image by the connected elements form an integral index, perform the primary and secondary searches, then the decision on line.

The features of the analogue method, which coincide with the features of the claimed technical solution, are as follows: recognition of a person’s face in the image.

The disadvantages of the known method and system are:

- the requirement of a priori knowledge of the error threshold, which is the allowable distance in the index space from the points in question to the point of the query index;

- the requirement of a priori face width for proportional image scaling.

A known system for identifying a person by facial image [Patent No. 2382408, IPC G06K 9/00]. The invention relates to the field of identification of a person by facial image. The technical result consists in accelerating the procedure and increasing the reliability of human identification. This result is achieved by the fact that the video image with the face of the person being identified is obtained from a frontal perspective; fragments with the image of the eyes are distinguished by preliminary selection of primary feature points corresponding to the location of the eyes, and comparison with predetermined standards of the right and left eye; the selection of informationally significant areas in the selected fragment of the information field is carried out using the Gauss function in the ellipse; building standards of faces of identifiable people is performed using at least two types of orthogonal basis functions for preliminary search and recognition, for detailed recognition and construction of additional standards for final recognition; identify a person by an integrated assessment of similarity measures for each analyzed fragment due to a two-stage procedure for comparing standards for preliminary search and recognition and comparison of standards for detailed and final recognition.

The features of the analog system, which coincide with the features of the claimed technical solution, are as follows: face recognition of a person and his search in the image.

The disadvantages of the known system are: the need for prior training of the system.

A known method of detecting faces in an image using a cascade of classifiers [Patent No. 2427911, IPC G06K 9/62]. The invention relates to the field of digital image processing, searches for randomly oriented persons in digital images. The technical result of the invention is to increase the speed and accuracy of the processed data when detecting faces in the image. The technical result is achieved due to the fact that in the method of detecting faces in the image, which includes the use of a cascade of homogeneous classifiers, each classifier is used to separate the input vector into two classes “face”, “non-face”, and each classifier is characterized by a higher classification accuracy compared to with the previous classifier in the cascade, they search for circles in the input image using the Hough transform for circles, form a set of fragments in the image areas In the case where circles were found, by dividing the areas into a set of fragments of the same size, apply a cascade of classifiers to each fragment and determine whether it contains a face or not, and to make a decision on whether or not a face exists, all classifiers must include the corresponding input vector constructed from the fragment, to the face class.

The features of the analogue method, which coincide with the features of the claimed technical solution, are as follows: finding arbitrarily oriented persons in digital images.

The disadvantages of this method are: the need for prior training of the system, does not produce identification. A known system for face recognition based on a list of people not subject to verification [Patent No. 2381553, IPC G06K 9/00, G06K 9/62]. The invention relates to biometric identification systems for the image of a person's face. The technical result is personality recognition in conditions of uneven lighting and for a wide range of angles of rotation of the face in the image. In the method, a gallery of images of “negative” persons is formed — those people who are not subject to verification, a gallery of famous people is formed with a face image converted into an internal representation adopted for recognition algorithms implemented in the method, a person’s face is detected in the input image, and a segmented region is prepared containing the face to recognition by normalizing the position of the face in accordance with the position of the eyes, scaling to a given size, normalizing the intensity In fact, they perform basic facial recognition algorithms that can calculate the similarity measures of two images, select a subset of “negative” faces, on the basis of which they determine the choice of a new recognition algorithm, form a new recognition algorithm based on converting proximity measures of the basic algorithm taking into account the generated gallery of “negative” faces carry out the classification according to the set of registered images.

The features of the analog system, which coincide with the features of the claimed technical solution, are as follows: recognition of a person’s face in the image.

The disadvantages of the known system are: the need to select a subset of "negative" persons.

Closest to the invention is a device for recognizing facial relief [Patent No. 2431190, IPC G06K 9/00].

Consider a prototype device involves:

- the formation of two images of the object, while one of the images is formed when the backlight is on, and the second when it is off;

- detection on each of the two images of the object areas containing faces, determining the coordinates of areas containing faces;

- comparing the detected areas, cutting out the area of the source images in accordance with the coordinates of the areas containing the faces, normalizing, determining the coordinates of the characteristic points of the face, aligning the cut areas with respect to each other so that the coordinates of the corresponding characteristic points coincide;

- Providence analysis of changes in pixel intensities of the mapped areas, the construction of a map of the distribution of intensities;

- formation of a decision on the relief represented on the face images.

For the functioning of the device, it is necessary to find at least one pair of characteristic points selected from a set of characteristic points containing centers and corners of the eyes, wings of the nose, corners of the mouth.

For the device to function, it is necessary to analyze the changes in the intensity of the pixels of the mapped areas, while comparing the intensities of the pixels in the aligned cut-out areas and calculate the vector of statistical estimates of the changes in the intensities of the corresponding sections of these areas.

For the device to function, it is necessary to formulate a decision on the relief of the face in the image, while the binary decisive function was applied to the vector of statistical estimates.

Consider a prototype device contains:

- backlight;

- means of imaging an object;

- means for detecting areas containing faces, configured to determine the coordinates of areas containing faces;

- a means of comparing the detected areas, made with the possibility of forming a matching detected areas by cutting out the found areas from the source images, normalizing, determining the characteristic points of the face and aligning the regions relative to each other according to the coordinates of the corresponding characteristic points;

- a tool for analyzing changes in the intensity of pixels of the mapped areas of the formed images, configured to build intensity distribution maps;

- a means of forming a decision on the relief represented on the face images.

The features of the prototype device that match the features of the claimed technical solution are as follows: recognition of a person's face; identification of a person by the image of his face.

The disadvantages of the known device are:

1. the selected area of the face may contain fragments of the background, which leads to recognition errors

2. The set of characteristic points of the image of the face may not contain individual features of the face.

The block diagram of a device that implements the considered algorithm includes an image capturing unit, a unit for searching for areas containing faces, a unit for selecting and comparing areas, a unit for analyzing changes in areas, and a decision generating unit.

The proposed device for automated face recognition when processing a group photo allows you to solve the problem of recognizing a group of faces in images with a textured background. The device implements the following algorithm:

1) receiving a stream of group photographs I containing faces of several people;

2) obtaining a test image containing the face of an identifiable person;

3) localization of faces in group photos from stream I using a Viola-Jones face detector;

4) the allocation of localized faces in group photos from stream I using the algorithm for dividing the image into background and object ('matting');

5) normalization of the received images, each image contains a person’s face on a black background;

6) the formation of a vector of signs of the obtained images of faces and test images using the method of principal components. The essence of the principal component method is to form the matrix Xm, which consists of all examples of facial images of the training set (vectors X). The covariance matrix Σ of the set Xm is decomposed into eigenvectors and eigenvalues:

Λ = С ^T · Σ · С, (1)

where C is the matrix of eigenvectors, Λ is the diagonal matrix of eigenvalues. From C a submatrix C _M is selected, corresponding to M the largest eigenvalues of the matrix Λ. Conversion:

$$Y=C_M^T\tilde{X},\text{(2)}$$

- нормализованный вектор с нулевым математическим ожиданием), характеризует большую часть общей дисперсии и показывает самые значительные изменения Xm. Выбор первых М главных компонент разбивает векторное пространство на главное пространство $$F={\left\{C_i\right\}}_{i=1}^M,$$

содержащее главные компоненты, и его ортогональное дополнение $$F={\left\{C_i\right\}}_{i=M+1}^M.$$

(Where $$\tilde{X}=X-\overline{X}$$

is a normalized vector with zero expectation), characterizes most of the total variance and shows the most significant changes in Xm. The choice of the first M principal components divides the vector space into the main space $$F={\left\{C_i\right\}}_{i=\mathrm{one}}^M,$$

containing main components, and its orthogonal complement $$F={\left\{C_i\right\}}_{i=M+\mathrm{one}}^M.$$

7) comparison of the feature vector of the tested face image with the feature vectors of all other face images in group photos from stream I based on the Euclidean metric.

The device for automated face recognition during group photo processing (Fig. 1) contains a block for localizing face regions 1, a delay unit 7, a control unit 9, the inputs of which are information inputs of the device. The output of the face region localization unit 1 is connected to the input of the image separation unit for the background and the object 2, the output of which is connected to the first input of the image normalization unit 3. The output of the control unit 9 is connected to the second input of the image normalization unit 3, the output of which is connected to the input of the vector forming unit features of faces 4, the output of which is connected to the input of the block of buffer 5, the output of which is connected to the first input of the comparison unit 6. The output of the delay unit 7 is connected to the input of the unit for forming the vectors of signs of face 8, the output is cat It is connected to the second input of the comparison unit 6, the output of which is the information output of the device.

The automated face recognition device when processing a group photo works as follows. The input stream of the localization unit 1 receives a stream of images on which the faces of several people are represented. At the same time, an image with the face of the person being identified is received at the input of the delay unit 7 and the control unit 9. Using the block for dividing the image into background and object 2, the regions of faces in the images from the stream are selected. In the normalization block 3, the selected areas of the faces are scaled to the image of the identifiable person. Information about the resolution of the identifiable person is supplied from the control unit 9 to the second input of the normalization block 3. In the units for generating facial features vectors 4 and 8, using the principal component method, eigenvectors and eigenvalues of the selected faces and faces of the identifiable person are formed. In block buffer 5, feature vectors of selected areas of faces are stored. In comparison block 6, the feature vectors of the selected areas of the faces are compared with the face features of the identifiable person's face vector.

EFFECT: automated face recognition during group photo processing.

Claims (1)

An automated face recognition device when processing a group photo, containing a unit for searching for areas containing faces, a decision-making unit, characterized in that the output of the face area localization unit (1) is connected to the input of the background image separation unit and the object (2), the output of which is connected to the first input of the image normalization block (3); the output of the control unit (9) is connected to the second input of the image normalization unit (3), the output of which is connected to the input of the face feature vector generation unit (4), the output of which is connected to the input of the buffer unit (5), the output of which is connected to the first input of the comparison unit (6); the output of the delay unit (7) is connected to the input of the face vector formation unit (8), the output of which is connected to the second input of the comparison unit (6), the output of which is the information output of the device.