KR20190107480A - Aparatus and method for face recognition - Google Patents

Abstract

Disclosed is a face recognition method. The face recognition method includes the following steps of: classifying one input face image into a first input face image divided by a first block unit and a second input face image divided by a second block unit; calculating first block similarity by comparing characteristic values of a first registration face image and the first input face image; calculating second block similarity by comparing characteristic values of a second registration face image and the second input face image; calculating final similarity between the registration face images and the input face images by adding up the first block similarity and the second block similarity; and extracting the final similarity in regard to every registration face image registered in the face recognition database, and outputting a registration face image with the largest final similarity as a result of recognizing the input face images. Therefore, the face recognition method is capable of preventing a deterioration in the performance of face recognition.

Description

Facial recognition device and method {APARATUS AND METHOD FOR FACE RECOGNITION}

The present invention relates to a face recognition technology, and more particularly, to a technology for determining whether a face image registered in advance and a face image currently input are the same person.

Face recognition is a technology for recognizing that a person is the same by extracting a feature from a face image and comparing face data learned through various classifiers with feature data extracted from a newly acquired face image.

For face recognition, feature data may be extracted from face images, such as LBP (Local Binary Features) extraction technique, Modified Census Transform (MCT) technique, and Gabor filter technique.

The features extracted from these methods are learned through classifiers. The classifiers used here are the Support Vector Machine, Artificial Neural Network, Nearest Neighbor Search, etc. have.

The face recognizer must be able to search newly acquired face images in a large scale in a pre-registered face database at a high speed. When a new face is added to the face database, it minimizes the update cost of the face database, It should be developed in a way that can be applied immediately.

For this reason, classifiers using nearest neighbor search are more commonly used than classifiers using support vector machines or neural networks.

In general, the nearest neighbor search in face recognition is a method of calculating face similarity by dividing two compared face images into blocks having the same ratio and comparing the features shown in both blocks at the same position.

However, the method based on the search for the nearest neighbors, if the block containing the components (noise) that does not help classification in the extracted feature is included in a high proportion, the similarity value between the two faces is lowered even though they are the same person. do. For example, when the input face image compared to the registered face image occurs in facial expression change, wearing glasses, and hair style change, the face recognition rate decreases.

In order to solve the problems of the related art, the present invention improves face recognition rate by preventing face similarity calculation errors by adaptively comparing features appearing in a multi-block position region without modifying feature descriptors or classifier learning. An object of the present invention is to provide a face recognition apparatus and method for providing the same.

In accordance with an aspect of the present invention, there is provided a face recognition method, including: a first registered face image in which a feature value is extracted from one registered face image in a first block unit, and a second extracted feature value in a second block unit; Dividing the registered face image into a face recognition database; One input face image is classified into a first input face image partitioned by the first block unit and a second input face image partitioned by the second block unit, and the first input face image and the second input face image. Extracting feature values from each block of; Comparing feature values of the first input face image and the first registered face image in block units, a first block similarity is calculated, and feature values of the second input face image and the second registered face image in block units. Calculating a second block similarity by comparison; Calculating a final similarity between the input face image and the registered face image by using the first block similarity and the second block similarity; And extracting the final similarity for all registered face images registered in the face recognition database, and outputting the registered face image having the maximum similarity as the recognition result of the input face image.

A face recognition apparatus according to an aspect of the present invention is a memory storing a face recognition database in which a plurality of registered face images are registered, wherein each registered face image includes a first registered face image and a first registered face image from which feature values are extracted in units of first blocks. The memory including a second registered face image from which feature values are extracted in units of two blocks; And classifying one input face image into a first input face image partitioned by the first block unit and a second input face image partitioned by the second block unit, wherein the first input face image and the second input face A feature value is extracted from each block of the image, and a first block similarity is calculated by comparing the feature values of the first input face image and the first registered face image on a block basis, and the second input face image and the first image are calculated. 2 a second block similarity is calculated by comparing feature values of a registered face image in units of blocks, and a final similarity between the input face image and the registered face image is calculated using the first block similarity and the second block similarity. And extracting the final similarity for all registered face images registered in the face recognition database, and inputting the registered face image having the maximum final similarity. And a processor for outputting a recognition result of the cave image.

According to the present invention, by performing the face recognition process through the adaptive block comparison (or matching) method according to the size of the block, it is possible to prevent the degradation of the face recognition performance due to the noise component of the input face image.

1 is a block diagram of a face recognition apparatus according to an exemplary embodiment.
2 is a diagram illustrating a registration face image and an input face image partitioned into different block units according to an exemplary embodiment.
3 is a flowchart of a face recognition method according to an exemplary embodiment.
4 is a detailed flowchart of step S310 shown in FIG.
FIG. 5 is a detailed flowchart of step S360 shown in FIG. 3.

Hereinafter, various embodiments of the present invention will be described in connection with the accompanying drawings.

Various embodiments of the present invention may have various changes and various embodiments, and specific embodiments are illustrated in the drawings and related detailed descriptions are described. However, this is not intended to limit the various embodiments of the present invention to specific embodiments, it should be understood to include all modifications and / or equivalents and substitutes included in the spirit and scope of the various embodiments of the present invention. In the description of the drawings, similar reference numerals are used for similar elements.

Expressions such as "comprises" or "can include" as used in various embodiments of the present disclosure indicate the existence of a corresponding function, operation or component disclosed, and additional one or more functions, operations or It does not restrict the components. In addition, in various embodiments of the present invention, the terms "comprise" or "have" are intended to designate that the features, numbers, steps, operations, components, parts, or combinations thereof described in the specification exist. Or other features or numbers, steps, operations, components, parts or combinations thereof in any way should not be excluded in advance.

In various embodiments of the present disclosure, the expression "or" includes any and all combinations of words listed together. For example, "A or B" may include A, may include B, or may include both A and B.

Expressions such as "first," "second," "first," or "second," and the like used in various embodiments of the present disclosure may modify various elements of the various embodiments, but do not limit the corresponding elements. Do not. For example, the above expressions do not limit the order and / or importance of the corresponding elements. The above expressions may be used to distinguish one component from another. For example, both a first user device and a second user device are user devices and represent different user devices. For example, without departing from the scope of the various embodiments of the present disclosure, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component.

When a component is said to be "connected" or "connected" to another component, the component may or may not be directly connected to or connected to the other component. It is to be understood that there may be new other components between the other components. On the other hand, when a component is referred to as being "directly connected" or "directly connected" to another component, it will be understood that there is no new other component between said one component and said other component. Should be able.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the embodiments. Singular expressions include plural expressions unless the context clearly indicates otherwise.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of the present invention belong.

The present invention can be applied to all kinds of systems for recognizing a face of a user by querying a face database previously established, determining whether the input face image of the user is a registered face image registered in the face database.

In particular, the present invention by recognizing the user's face through the block-based matching method (comparison method) adaptively changed the block size of the input face image and the registered face image, the noise component (expression) present in the input face image of the user Performance of facial recognition due to a change, wearing an accessory, or changing a hair style) can be prevented.

1 is a block diagram of a face recognition apparatus according to an embodiment of the present invention.

Referring to FIG. 1, the face recognition apparatus 100 may include an image acquirer 110, a processor 120, a user interface 130, a memory 140, a display 150, and a communicator 160. .

The image acquisition unit 110 acquires an input image input from a camera (not shown). Here, the camera (not shown) may include a depth recognition camera, a stereo camera and a color camera.

The input image includes a previous input image including the face of the user compared to the recognition target and a current input image including the face of the same user as the recognition target.

The previous input image is an image acquired at a previous time, and the current input image is an image obtained at the current time.

Due to factors such as a change in facial expression of a user, wearing of an accessory (eg, glasses), and a hair style, it is assumed that a face of a user included in a previous input image and a face of the same user included in a current input image are different.

The previous input image is processed into a registered face image to be compared with a face recognition target, and is registered in the face recognition database 140, and the current input image is processed into an input face image to be a face recognition target.

The processor 120 executes a face registration process of converting the input previous input image into a registered face image and registering it in the face recognition database 142.

When the face registration process is completed, the processor 120 converts the input current input image into an input face image (or a query face image), compares the input face image with the registered face image in units of blocks, and performs a face on the input face image. Run the recognition process. In this case, the processor 120 does not compare the input face image and the registered face image in units of blocks based on a fixed block size, but blocks the input face image and the registered face image based on a plurality of adaptively changed block sizes. Compare in units. This comparison is referred to as the "adaptive block comparison (or matching) method".

In order to execute the face registration process and the face recognition process, the processor 120 may be implemented by, for example, a central processing unit (CPU), a graphic processing unit (GPU), or a combination thereof. Description of the processor 120 is described in detail below.

The user interface 130 may transmit a command input by the user to the processor 120. The command may, for example, relate to the start, control and end of the face registration process and the face recognition process. The user interface 130 may be various input means based on, for example, a keyboard and a touch screen.

The memory 140 may store a plurality of software required for executing the face registration process and the face recognition process. Furthermore, the memory 140 may store intermediate data, final result data, etc. generated by the face registration process and the face recognition process. The intermediate data may be data output from the components 121, 122, 123, 124, and 125 in the processor 120, and the final result data may be the face recognition database 142 and the face recognition unit in the processor 120. And a face recognition result output from 126).

Memory 140 includes non-volatile memory that does not require a continuous power supply to maintain stored information and volatile memory that requires a continuous power supply to maintain stored information. . Non-volatile memory includes, for example, a ROM, a flash memory, a hard disk, an optical disk drive, and the like. The volatile memory may be, for example, RAM.

The display 150 may display information about a process or a result of the face registration process and the face recognition process.

The communicator 160 transmits processing results according to the face registration process and the face recognition process to an external device or receives data information related to the face registration process and the face recognition process from an external device, and supports wired communication or wireless communication. It may be. Wireless communication includes mobile communication, short-range wireless communication, Internet communication, and the like, for example.

Hereinafter, the face registration process and the face recognition process executed by the processor 120 will be described in detail.

Processor 120

The processor 120 may be configured to include a plurality of units divided into functional units to execute the face registration process and the face recognition process. For example, the processor 120 may include a face detector 121, a preprocessor 122, a block partitioner 123, a feature value extractor 124, a similarity calculator 125, and a face recognizer 126. It may include.

In FIG. 1, dotted arrows and solid arrows appear between the components in the processor 120. The dashed arrows indicate the flow of signals that each component outputs according to the execution of the face registration process, and the solid arrows indicate the flow of signals that each component outputs according to the execution of the facial recognition process.

In the execution order of the processes, it should be noted that the face recognition process is executed after the face registration process is completed. Therefore, in each of the following configurations, the information processing process related to the face registration process (processing of the previous input image input at the previous time) and the information processing process related to the face recognition process (processing of the current input image input at the current time) are described together. However, it should be noted that this is for convenience of description only and is not executed at the same time.

Face Detector (121)

The face detector 121 outputs a face region detected from the input previous input image P1 as a face image P2 (hereinafter, a registered face image) and faces a face region detected from the input current input image C1. It detects as an image (C2, hereinafter, input face image).

As a face detection method, a knowledge-based method, a feature-based method, a template matching method, an appearance-based method, and the like may be used. As a specific example, a face from an input image using an Adaboost learning algorithm based on a Haar-like feature may be used. The area can be detected. However, the present specification is not limited to a specific face detection technique.

For reference, the AdaBoost learning algorithm is one of classification techniques for grouping strong classifiers with high detection performance through linear combination of several weak classifiers.

Preprocessor (122)

The preprocessor 122 performs a pre-processing process on the registered face image P2 and the input face image C2 input from the face detector 121, thereby preprocessing the registered face image P3 and the preprocess. The output input face image C3.

The preprocessor 122 resizes the registered face image P2 and the input face image C2, for example, and normalizes them, respectively, and normalizes the registered face image P2 and the input face image C2. For each pixel, preprocessing for converting a local binary pattern (LBP) pattern may be performed.

The binary pattern conversion method may use one of a Modified Center Symmetric Local Binary Pattern (MCS-LBP), a Modified Local Directional Pattern (MLDP), and a Centralized Gradient Pattern (CGP).

block Compartment (123)

The block partitioner 123 partitions the preprocessed registered face image P3 and the preprocessed input face image C3 into various block units in which the block size is adaptively changed.

In an example, the block partitioner 123 may partition the registered face image P3 and the input face image C3 in units of first blocks that define the first block size. An example of a registered face image and an input face image divided in units of first blocks is shown in FIG. 2.

In FIG. 2, P4-1 is a reference number indicating an example of a registered face image (hereinafter, referred to as a 'first registered face image') divided in a first block unit, and C4-1 is divided in a first block unit. Reference number indicating an example of an input face image (hereinafter referred to as a 'first input face image'). As illustrated in FIG. 2, in the case of the first registered face image and the first input face image partitioned by the first block unit, the block may be divided into 2 × 2 blocks.

In another example, the block partitioner 123 may partition the registered face image P3 and the input face image C3 in units of second blocks that define a second block size smaller than the first block size.

In FIG. 2, P4-2 is a reference number indicating an example of a registered face image (hereinafter, referred to as a 'second registered face image') divided in units of second blocks, and C4-2 is divided in units of second blocks. Reference number indicating an example of an input face image (hereinafter referred to as a 'second input face image'). As illustrated in FIG. 2, in the case of the second registered face image and the second input face image partitioned by the second block unit, the second registered face image may be divided into 4 × 4 blocks.

In another example, the block partitioner 123 may partition the registered face image P3 and the input face image C3 in units of third blocks that define a third block size smaller than the size of the second block.

In FIG. 2, P4-3 is a reference number indicating an example of a registered face image (hereinafter, referred to as a 'third registered face image') divided in units of third blocks, and C4-3 is divided in units of third blocks. Reference number indicating an example of an input face image (hereinafter referred to as a 'third input face image'). As illustrated in FIG. 2, in the case of the third registered face image and the third input face image partitioned by the third block unit, the third registered face image may be divided into 6 × 6 blocks.

Although not shown, in a further example, the block dividing unit 123 may divide the registered face image and the input face image into more blocks.

For convenience of explanation, hereinafter, it is assumed that the block dividing unit 123 divides the preprocessed registered face image P3 and the input face image C3 into first block units and second block units, respectively.

Feature Value Extraction unit (124)

The feature value extractor 124 extracts the feature value (s) from each block of the first registered face image P4-1 partitioned by the first block unit, and the second registered face image partitioned by the second block unit. The feature value (s) are extracted from each block of (P4-2). For example, when the first registered face image P4-1 is partitioned into 2 × 2 first block units, 2 × 2 feature values are extracted, and the second registered face image P4-2 is four. In the case of being partitioned into x 4 second block units, 4 x 4 feature values are extracted.

As described above, the first registered face image P5-1 from which the feature value is extracted in the first block unit and the second registered face image P5-2 from which the feature value is extracted in the second block unit are the face recognition database 142. Is registered).

The feature value extractor 124 extracts the feature value (s) from each block of the first input face image C4-1 partitioned by the first block unit, and the second input face image partitioned by the second block unit. The feature value (s) is extracted from each block of (C4-2). For example, when the first input face image P4-1 is partitioned in the first block unit, 2 × 2 feature values are extracted, and the second input face image P4-2 is in the second block unit. When partitioned, 4x4 feature values are extracted.

As described above, the first input face image C5-1 from which the feature value is extracted in the first block unit and the second input face image C5-2 from which the feature value is extracted in the second block unit are the face recognition database 142. In order to be compared with each of the first registered face image P5-1 and the second registered face image P5-2 registered in FIG. 9), the similarity calculator 125 is output to the similarity calculator 125.

On the other hand, the feature value extracted from each block refers to information representing the unique specification for the face. As a method of extracting such feature values, PCA (Principal Component Analysis), LDA (Linear Discriminate Analysis), etc. may be selectively used.

As another example, a 2-Dimensional Principal Component Analysis (2D-PCA) technique may be used to extract feature information of a face image or a converted image based on a two-dimensional image matrix.

Since the size of the projection matrix is smaller than that of the general PCA method, the 2D-PCA technique can be easily applied to portable devices, low-end embedded systems, etc., because the amount of computation and required memory can be relatively low.

Similarity Calculation (125)

The similarity calculator 125 calculates a first block similarity by comparing feature values of the first input face image and the first registered face image in block units, and calculates the second input face image and the second registered face image. The second block similarity is calculated by comparing the feature values in units of blocks.

Specifically, in order to calculate the first block similarity, the similarity calculator 125 equals the feature value of the first input face image C5-1 and the feature value of the first registered face image P5-1. A plurality of block similarities are calculated by comparing blocks of positions in block units. In this case, the feature value of the first input face image C5-1 partitioned into 2 × 2 blocks and the feature value of the first registered face image P5-1 partitioned into 2 × 2 blocks are blocked. When the similarity between the corresponding blocks is calculated by comparing in units, 2 × 2 block similarities may be calculated.

When the calculation of the plurality of block similarities is completed, the similarity calculator 125 determines the middle of the plurality of block similarities except the upper block similarities larger than an upper limit and the lower block similarities smaller than a lower limit according to a preset selection rule. Select block similarities. Here, the upper limit value and the lower limit value may be set in various ways according to the design. For example, the upper limit value may be the lowest block similarity among the block similarities belonging to the upper 25% among the plurality of block similarities, and the lower limit may be set to the plurality of block similarities. It may be the highest block similarity among the block similarities belonging to the lower 25% among the block similarities of.

When the selection of the intermediate block similarities is completed, the similarity calculator 125 calculates the first block similarity by adding the intermediate block similarities.

The calculation process of the second block similarity is similar to the calculation process of the first block similarity.

That is, the similarity calculator 125 calculates a plurality of block similarities by comparing feature values of the second input face image C5-2 and the second registered face image P5-2 in units of blocks. According to the selection rule set in the above, after selecting the remaining middle block similarities except the upper block similarity larger than the upper limit and the lower block similarity smaller than the lower limit among the plurality of block similarities, the selected intermediate block similarities are summed Calculate the two block similarity.

When the calculation of the first and second block similarities is completed, the similarity calculator 125 adds the first block similarity and the second block similarity, and thus the input face image C5 and the registered face image P5. Calculate the final similarity between the livers. This final similarity is calculated for all registered face images registered in the face recognition database.

As described above, the final similarity between the input face image C5 and the registered face image P5 is calculated by summing all the block similarities calculated for each block size, thereby increasing the accuracy of the final similarity.

If the block size is further subdivided to increase the number of block similarities, the more accurate final similarity may be calculated. However, in this case, since the number of block pairs to be compared increases, the amount of calculation according to the similarity calculation may increase. Therefore, in consideration of hardware performance, a range of variable block sizes needs to be appropriately adjusted.

The range of block sizes can be selected by the user. For example, the block partition unit 123 has five partitions each composed of 2 × 2 block units, 4 × 4 block units, 8 × 8 block units, 16 × 16 block units, and 32 × 32 block units. When supporting the process, three or four compartment processes may be executed according to a selection command input from the user through the user interface 130.

Meanwhile, the block similarity may be calculated by Equation 1 below.

는 L번째 등록된 사람의 g 번째 등록 얼굴 영상에서 i 번째 행, j 번째 열에 위치한 블록의 특징값이다. 따라서 상기 수학식 1은 입력 얼굴 영상과 등록 얼굴 영상을 소정의 블록 단위로 블록화시켜 블록 매트릭스를 구성하였을 때, 입력 얼굴 영상의 (i, j)째 블록과 등록 얼굴 영상의 (i, j)째 블록의 유사도를 산출하기 위한 식이다.Here, sim (a, b) is a function for calculating the similarity between a and b, F _input (i, j) is a feature value of the block located in the i th row and j th column of the input face image input by the query.

Is the feature value of the block located in the i th row and the j th column of the g th registration face image of the L th registered person. Therefore, in Equation 1, when the input face image and the registered face image are blocked by a predetermined block unit to form a block matrix, the (i, j) th block of the input face image and the (i, j) th of the registered face image The equation for calculating the similarity of blocks.

Face Recognition (125)

The face recognition unit 125 extracts the final similarity calculated by the similarity calculator 125 for all registered face images registered in the face recognition database 140 and inputs the registered face image having the maximum final similarity. Output as a result of face image recognition. That is, the face recognition unit 125 recognizes the registrant having the largest value among individual final similarity values registered in the face recognition database 140 as a person matching the input face image.

The recognition result may be output in the form of visual information through the display 150 or transmitted to an external device located at a far distance through the communication unit 160.

As described above, according to the face recognizing apparatus 100 of the present invention, by calculating the final similarity between the input face image C5 and the registered face image P5 by summing all the block similarities calculated for each block size, By increasing the accuracy of the final similarity, ultimately, errors in face recognition can be minimized.

3 is a flowchart of a face recognition method according to an embodiment of the present disclosure, FIG. 4 is a detailed flowchart of step S310 illustrated in FIG. 3, and FIG. 5 is a detailed flowchart of step S360 illustrated in FIG. 3. In the steps below, contents overlapping with those described with reference to FIGS. 1 and 2 will be briefly described or omitted.

Referring to FIG. 3, in step S310, a process of registering a face registration image from which feature values have been extracted into a face recognition database is performed. A detailed process for this is shown in FIG. 4.

Referring to FIG. 4, in order to register a face registration image in a face recognition database, first, in step S311, a face region P2, hereinafter, a registered face image from a previous input image P1 obtained from the image obtaining unit 110. ) Is performed. Here, as a method of detecting the registered face image P2, a knowledge-based method, a feature-based method, a template matching method, an appearance-based method, a Haar-like feature-based Adaboost learning algorithm, and the like may be used.

Next, in step S313, a process of preprocessing the detected registered face image P2 is performed. For example, the preprocessing may be performed by resizing the registered face image P2 to normalize it, and converting a LBP pattern for each pixel of the normalized registered face image P2. Can be. Here, one of MCS-LBP (Modified Center Symmetric Local Binary Pattern), MLDP (Modified Local Directional Pattern), and CGP (Centralized Gradient Pattern) may be used as the binary pattern conversion method.

Subsequently, in step S315, a process of dividing the preprocessed registered face image P3 into different block units and classifying the first and second registered face images P4-1 and P4-2 is performed. For example, the first registered face image P4-1 divided into pre-processed registered face images P3 by a first block unit in which one block is defined as a first block size, and one block are the first blocks. The image is classified into a second registered face image P4-2 partitioned into second block units defined by a second block size smaller than the size. In this case, the number of blocks partitioned in the first registered face image P4-1 is M × N, and the number of blocks partitioned in the second registered face image P4-2 is m × n. Here, M × N is a value smaller than m × n.

Next, in step S317, a feature value is extracted from each block of the first registered face image P4-1, and a feature value is extracted from each block of the second registered face image P4-2. Since the number of blocks partitioned in the first registered face image P4-1 is M × N, the feature value extracted from the first registered face image P4-1 includes M × N feature values. Since the number of blocks partitioned from the two registered face images P4-2 is m × n, the feature value extracted from the second registered face image P4-2 includes m × n feature values.

In operation S319, the first registered face image P5-1 from which the feature values (M × N feature values) are extracted and the second registered face image from which the feature values (m × n feature values) are extracted A process of registering (P5-2) to the face recognition database 142 is performed.

Referring back to FIG. 3, in step S320, a process of detecting an input face image C2 as a recognition target from the current input image C1 is performed. Since the detection process of the input face image C2 is the same as the detection process S311 of the registered face image, the description thereof will be omitted.

Subsequently, in step S330, a process of preprocessing the detected input face image C2 is performed. Since the preprocessing process of the input face image C2 is the same as the preprocessing process S313 of the registered face image P2, a description thereof will be omitted.

In operation S340, a process of dividing the preprocessed input face image into different block units and classifying the first and second input face images C4-1 and C4-2 is performed. For example, a first input face image C4-1 divided into a first block unit in which one block is defined as a first block size, and one block are preprocessed input face images C3. The image is classified into a second input face image C4-2 partitioned into second block units defined by a second block size smaller than the size. In this case, the number of blocks partitioned in the first input face image C4-1 is M × N, and the number of blocks partitioned in the second input face image C4-2 is m × n. Therefore, the number of blocks partitioned in the first input face image C4-1 and the number of blocks partitioned in the first registered face image P4-1 are the same, and the number of blocks partitioned in the first input face image C4-2 is the same. The number of partitioned blocks and the number of blocks partitioned in the second registered face image P4-2 are the same.

In operation S350, a feature value (M × N feature values) is extracted from each block of the first input face image P4-1, and the feature is extracted from each block of the second input face image P4-2. A process of extracting a value (m × n feature values) is performed. Here, the method of extracting the feature value is the same as the method of extracting the feature value from each block of the registered face image.

Subsequently, in operation S360, a process of calculating a final similarity is performed by comparing the feature values of the first and second input face images and the first and second face registered images in units of blocks. A detailed process for this is shown in FIG. 5.

Referring to FIG. 5, in order to calculate a final similarity, first, in step S361, a feature value (M × N feature values) of the first input face image C5-1 and the first registered face image ( P5-1) compares the feature values (M × N feature values) in units of blocks (comparing 1: 1 blocks of the same position) to calculate a plurality of (M × N) block similarities. Is performed.

Subsequently, in step S362, a process of selecting intermediate block similarities among the calculated (M × N) block similarities is performed. For example, the middle block similarities other than the upper block similarities larger than the upper limit and the lower block similarities smaller than the lower limit are selected from among the plurality (M × N) block similarities according to a preset selection rule. Here, the upper limit value and the lower limit value may be set in various ways according to the design. For example, the upper limit value may be the lowest block similarity among the block similarities belonging to the upper 25% among the plurality of block similarities, and the lower limit may be set to the plurality of block similarities. It may be the highest block similarity among the block similarities belonging to the lower 25% among the block similarities of.

Then, in step S363, the process of calculating the first block similarity is performed by summing the selected intermediate block similarities.

Thereafter, a calculation process of the second block similarity is performed. The calculation process of the second block similarity is the same as the calculation process of the first block similarity.

Specifically, in step S364, the feature values (m × n feature values) of the second input face image C5-2 and the feature values (m × n number) of the second registered face image P5-2. The feature values) are compared in units of blocks (comparing 1: 1 blocks of the same position) to calculate a plurality (m × n) of block similarities.

Subsequently, in step S365, a process of selecting middle block similarities among the calculated (m × n) block similarities is performed. For example, the middle block similarities except the upper block similarities larger than the upper limit and the lower block similarities smaller than the lower limit are selected from the plurality (m × n) block similarities.

Subsequently, in operation S366, a process of calculating the second block similarity is performed by summing the selected intermediate block similarities.

When the first and second block similarities are calculated, a process of calculating the final similarity is performed by summing the first block similarity and the second block similarity.

Referring again to FIG. 3, when the calculation of the final similarity is completed, in step S370, the final similarity is calculated for all registered face images registered in the face recognition database 142, and the registered face image having the maximum final similarity is maximum. After the process of outputting the result of the recognition as the result of the input face image, all the series of processes related to face recognition of the input face image are completed.

Although the present invention has been described above with reference to the embodiments, these are merely examples and are not intended to limit the present invention. It will be appreciated that various modifications and applications are not illustrated. For example, each component shown in detail in the embodiment of the present invention may be modified. And differences relating to such modifications and applications will have to be construed as being included in the scope of the invention defined in the appended claims.

Claims (15)

Classifying one registered face image into a first registered face image from which feature values are extracted in units of first blocks and a second registered face image from which feature values are extracted in units of second blocks, and registering the registered face images in a face recognition database;
Classifying one input face image to be recognized into a first input face image from which feature values are extracted in units of the first block and a second input face image from which feature values are extracted in units of the second block;
Comparing feature values of the first input face image and the first registered face image in block units, a first block similarity is calculated, and feature values of the second input face image and the second registered face image in block units. Calculating a second block similarity by comparison, and calculating a final similarity between the input face image and the registered face image by using the first block similarity and the second block similarity; And
Extracting the final similarity for all registered face images registered in the face recognition database, and outputting a registered face image having the maximum similarity as a recognition result of the input face image.
The face recognition method of claim 1, wherein the size of one block partitioned by the first block unit and the size of one block partitioned by the second block unit are different from each other.
The method of claim 1, wherein the registering comprises:
Classifying the one registered face image into the first registered face image partitioned into M × N first block units and the second registered face image partitioned into m × n second block units;
Extracting a feature value from each block of the first registered face image and extracting a feature value from each block of the second registered face image using a feature extraction technique; And
Registering a first registered face image and a second registered face image from which feature values have been extracted into the face recognition database;
Face recognition method comprising a.
The method of claim 1, wherein the extracting comprises:
Classifying one input face image into a first input face image partitioned into M × N first block units and a second input face image partitioned into m × n second block units; And
Extracting feature values from each block of the first input face image and the second input face image using a feature extraction technique;
Face recognition method comprising a.
The method of claim 1, wherein the calculating
Calculating M × N block similarities by comparing feature values of the first input face image and the first registered face image in units of blocks;
Selecting middle block similarities except for upper block similarities larger than an upper limit and lower block similarities smaller than a lower limit among the M × N block similarities according to a preset selection rule;
Calculating the first block similarity by summing the intermediate block similarities selected from the M × N block similarities;
Calculating m × n block similarities by comparing feature values of the second input face image and the second registered face image in units of blocks;
Selecting middle block similarities except for upper block similarities larger than an upper limit and lower block similarities smaller than a lower limit among the m × n block similarities according to a preset selection rule;
Calculating the second block similarity by summing the intermediate block similarities selected from the m × n block similarities.
Face recognition method comprising a.
The method of claim 1, wherein the calculating of the final similarity degree comprises:
And calculating the final similarity by adding the first block similarity and the second block similarity.
A memory storing a face recognition database in which a plurality of registered face images are registered, wherein each registered face image includes a first registered face image from which feature values are extracted in units of first blocks and a second extracted feature value in units of second blocks. The memory including a registered face image; And
Classify one input face image into a first input face image from which feature values are extracted in units of the first block, and a second input face image from which feature values are extracted in units of the second block. A first block similarity is calculated by comparing feature values of a registered face image in block units, and a second block similarity is calculated by comparing feature values of the second input face image and a second registered face image in block units. Calculate a final similarity between the input face image and the registered face image by using the first block similarity and the second block similarity, and extract the final similarity for all registered face images registered in the face recognition database. And a processor configured to output a registered face image having the maximum final similarity as a recognition result of the input face image.
Face recognition device comprising a.
The method of claim 7, wherein the processor,
Prior to performing a face recognition process on the one input face image, the first registered face image and m × n pieces of the first registered face image obtained by extracting feature values of each registered face image in units of the M × N blocks; And classifying a second registered face image from which feature values have been extracted in units of two blocks, constructing the face recognition database, and storing the constructed face recognition database in the memory.
The method of claim 7, wherein the processor,
Classifying the one input face image into a first input face image partitioned into M × N first block units and a second input face image partitioned into m × n second block units, and extracting a feature extraction algorithm. And extracting a feature value from each block of the first input face image and extracting a feature value from each block of the second input face image.
The method of claim 7, wherein the processor,
M × N block similarities are calculated by comparing the feature values of the first input face image and the first registered face image on a block basis, and an upper limit value is calculated among the M × N block similarities according to a preset selection rule. A process of selecting the intermediate block similarities except for larger upper block similarities and lower block similarities smaller than a lower limit, and summing the intermediate block similarities selected from the M × N block similarities to calculate the first block similarity Face recognition device that is to run.
The method of claim 7, wherein the processor,
By comparing feature values of the second input face image and the second registered face image in units of blocks, m × n block similarities are calculated, and an upper limit value among the m × n block similarities according to a preset selection rule. A process of selecting the middle block similarities except for larger upper block similarities and lower block similarities smaller than a lower limit, and summing the middle block similarities selected from the m × n block similarities to calculate the second block similarity Face recognition device that is to run.
The method of claim 7, wherein the processor,
And summing the first block similarity and the second block similarity to calculate the final similarity.
The apparatus of claim 7, wherein each of the face registration images stored in the memory comprises:
And a third registered face image from which feature values are extracted in units of third blocks.
The processor of claim 13, wherein the processor comprises:
And classifying one input face image into a first input face image, a second input face image, and a third input face image divided into units of the third block.
The processor of claim 14, wherein the processor comprises:
Extracting a feature value from each block of the third input face image;
A process of further calculating a third block similarity by comparing feature values of the third input face image and the third registered face image on a block basis;
A process of calculating a final similarity between the input face image and the registered face image by summing up the first block similarity, the second block similarity, and the third block similarity;
Face recognition device that is to run.

Images