KR20100091516A - Method for identifying image face and system thereof - Google Patents

Abstract

PURPOSE: A method for identifying an image face and a system thereof are provided to accurately identify a face of a person even in a circumference where an external additional object exists and accurately perform the identification of the person according to the monitoring list. CONSTITUTION: A face detecting block(11) defects a face from an image of a person obtained by a camera, and an image preprocessing block(12) preprocesses a face image. A coding block(13) makes and processes a feature point map for the preprocessed image and converts the image into a compact representation format. A template comparing block(15) determines the grade of an image by measuring the similarity of an image of the representation and the registered image.

Description

Method for identifying image face and system approximately}

The present invention relates to an image face identification method, and more particularly, to a method and a system for identifying a person.

Identifying people based on facial images has been an important topic in recent decades. The system for identifying a face image of a person is based on comparing a similar model obtained according to an input image with a face model (or representation) of a specific person already known. Not so long ago, in most similar systems, the images to be compared had to be images of the person's front face obtained under specific lighting conditions. These requirements have limited the practical application of the recognition system.

Several methods have been proposed to solve these problems, and such approaches are described in the following patents: US Pat. No. 5,642,431, Patent No. 6,108,437, Patent No. 6,775,411, Patent No. 6,944,319, Patent No. 7,127,087 , Patent number 7,142,697.

These prior arts have proposed face identification methods that can identify various aspects of the face instead of the front of the person in uneven lighting conditions, where the invariability of the lighting conditions is secured by processing the original image. However, this method lacks the reliability and stability of work in low light and obstructive conditions.

US Pat. No. 6,826,300 of the prior art discloses a method of measuring proximity to a standard (template) image. Here, a method of manufacturing a face image feature point map based on the Gabor wavelet basis is used, and based on the main component and the linear identification analysis method and performing proximity measurement, an effective feature point of the face image is selected. At this time, the feature map generated to distinguish the most unique and firm facial features is not processed.

However, as with other similar and prototypical methods, there is still a problem of low precision and stability when working in conditions of varying lighting, face position for various cameras, and obstructions.

It is an object of the present invention to provide a person identification method and system for recognizing an image of a specific person more accurately.

Another object of the present invention is to provide a method and a system for improving the identification accuracy of a face even under a condition in which there is an obstacle, an uneven illumination, and an external subject such as glasses.

Another object of the present invention is to provide a method and system for accurately identifying a person in consideration of features of persons in a special list even under uneven illumination and wide rotation angle bands on an image.

For the above problem, the person identification method according to the features of the present invention comprises the steps of forming a feature point map of the image to be tested; Measuring a similarity between a feature point map of the image and registered images already processed and registered in a database; And determining the grade of the image under test based on the result of the similarity measurement.

According to another aspect of the present invention, a person identification system includes a face detection block that detects a face from an image of a person obtained by a camera; An image preprocessing block which performs preprocessing on the detected face image; A coding block for creating a feature point map for the preprocessed image and processing the feature point map into a set compact representation; A registration image database storing registration images for the subjects; And a template comparison block that measures similarity between the registered image and the image according to the obtained expression and determines the grade of the image based on the similarity.

According to the present invention, it is possible to accurately identify an image of a specific person. In particular, even in the presence of noise, shade, and additional objects (glasses, etc.), face identification of a person can be accurately performed. In addition, the identification of the person according to the watch list (the watch list) is accurate.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification, when a part is said to "include" a certain component, it means that it can further include other components, without excluding other components unless specifically stated otherwise. In addition, terms such as "... block", "... module" described in the specification means a unit for processing at least one function or operation, which may be implemented in hardware or software or a combination of hardware and software.

In the person identification method according to an embodiment of the present invention, the following operations are performed.

-Form a feature point map of the registered image

Convert the set of matrix values in the feature point map of the image to a matrix with predefined values

A plurality of registered images are generated, each of which is based on a subject to be recognized.

-Apply the image feature point map set classification algorithm, which is capable of calculating two sets of similarity measurement, determine the identity, evaluate the proximity level of each of the image under test and the registered image, and classify the image under test .

In order to achieve an optimal effect in this method, a registered image feature map may be formed by convolutioning the cores of the image matrix and the gradient oriented filters.

For the best effect in this method, we can regard the core of the staged directivity filter as complex and use the absolute values of the convolution as a feature map of the registered image.

In this method, for optimal effect, the gradual directional filter core can be regarded as complex, but the phases of the convolution result can be used as the feature point map of the registered image.

In this method, for an optimal effect, an even distribution may be derived by a feature map in which the filtering results are arranged completely or partially.

Meanwhile, the person identification system according to an embodiment of the present invention may be implemented as an apparatus including at least one processor and the following elements.

A database that can provide templates to compare templates and make decisions

Tools for creating a feature point map of an image and a tool for converting values of a matrix corresponding to the feature point map of an image into a matrix having a predetermined distribution of values.

A basic image classifier implemented to measure the similarity of two images and to make perceptual decisions based on the results

Now, with reference to the accompanying drawings, a person identification method and system according to an embodiment of the present invention will be described.

1 is a structural diagram of a person identification system according to an embodiment of the present invention.

As illustrated in the accompanying FIG. 1, the person identification system according to an exemplary embodiment of the present disclosure may include a camera 10 for capturing an image of a person and a face detection block 11 for detecting a face from an image of the person. ), The image preprocessing block 12 which performs preprocessing on the detected face image, and coding to create a contour map for the preprocessed image and convert it into a compact representation according to an embodiment of the present invention. Block 13, a registered image database 14 storing registered images of subjects to be recognized, and measuring similarity between the registered image and the image according to the acquired representation system, and determining the classification of the corresponding image based on this. The template comparison block 15 is included.

The registration image database 14 may provide data for comparing templates and making decisions.

The coding block 13 may include tools for producing a feature point map of the image, and may also include a tool for converting values of a matrix corresponding to the feature point map of the image into a matrix having a distribution of predetermined values. .

The template comparison block 15 may include a basic image classifier that is implemented to measure similarity for two images and to determine recognition based on the results.

Next, a person identification method according to an embodiment of the present invention will be described based on the structure.

Person identification method according to an embodiment of the present invention is based on a technology that can effectively solve the following problems.

Further processing of the data to be recognized

-Improved stability for lighting conditions and camera angles of people presented in the encryption stage just before classification

To produce a distribution histogram based on some form of stepwise directional filtering at the coding stage of the template image and the test image by converting the processing result into a digital format, that is, a ranking matrix.

2 is a flowchart illustrating a person identification process according to an exemplary embodiment of the present invention, and in particular, a flowchart illustrating an identification process performed in association with each component of the person identification system.

First, the camera 10 acquires a two-dimensional image of the subject OB (S100), and transfers the acquired subject image to the face detection block 11 and the image preprocessing block 12 (S110). The face detection block 11 detects a face of a person from an input subject image (S120), and extracts face coordinates that are information about the detected face, that is, the position and size of a portion corresponding to the detected face in the subject image. Transfer to 12 (S130).

The image preprocessing block 12 recognizes a face part in the input subject image based on face coordinates transmitted from the face detection block 11 (S140), and cuts and scales a face part with respect to the recognized face part. ), And performs a preprocessing process including the conversion to the color (S150).

The preprocessed image is transferred to the coding block 13, and the coding block 13 configures a template as follows (S160).

First, a feature point map for an image is produced (S161). In this case, a feature point map is generated based on a set of gradient oriented filters, and a Gabor wavelet may be used as the gradient directive filter. In this case, the comparison between the input image and the feature point map set is a filtering result, and each of the feature point maps in this set is a convolution of the Gabor filter core with a constant direction and frequency.

Next, a procedure of substituting a value corresponding to a predetermined distribution to each element of the obtained feature point map is performed (S162). For example, substituting values with an even distribution in cutting [-1,1] (leveling of distribution histograms). Each element of the feature map is compared with a unique integer-valued index from integers 0 through K-1, where K represents the number of elements that make up the feature map. It should not be less than the index of any other element with a small value.

In this way, the feature map is compared with a matrix with a corresponding magnitude class. In order to obtain the substituted feature point map, matrix elements of each class are divided by K-1 and an inverse function of the required distribution is calculated based on the obtained values (S163). This process is described in more detail in FIG.

3 is an exemplary view illustrating a feature point map manufacturing process described above. Specifically, as shown in FIG. 3, the preprocessed image 30, such as a ratio adjustment, is convolved with the directional filter 31, resulting in a feature point map 33 comprising elements of M × M. The feature point map 33 is then classified 34 to obtain a matrix 35 having a corresponding magnitude class, where each element of the matrix is an integer from 1 to M × N−1.

As described above, an inverse graph of a predetermined distribution is obtained based on a value obtained by dividing each matrix element by K-1, that is, M × N −1 (36). In the obtained inverse function graph 36 of the required distribution, it is shown that each element of the matrix 35 and the feature point map 37 composed of M × N coincide with each other, and the histogram is presented in a constant form.

As described above, an inverse function of a predetermined distribution of a feature point map is calculated, and the above process is performed on multiple face images of the same person obtained at different conditions and at different times, thereby performing a plurality of feature point maps. Obtain a feature point map set consisting of:

Next, the feature point map set is represented compactly (S164). In this case, a linear discrimination analysis method, that is, a linear discrimination analysis method having a characteristic of minimizing the variation in the class when the variation between the main component method and the classes is minimum, may be sequentially used to express the feature point map set compactly. Can be. Here, a class is a set of multiple facial feature maps of the same person, all obtained at different conditions and at different times. The recognition algorithm establishes a basis in the initial space of the feature point map set by the above-described processes based on a learning sample during the curriculum, and the basis optimizes the combination of feature map set classes ( In the algorithm or algorithms used) (character space). This set can be fold-out or resolved in the feature space for compact representation of the feature point map set introduced in the recognition process.

Meanwhile, the feature point maps of the registered images are also produced through the above-described method, and the feature point maps of the registered images are stored in the registered image database 14 to be used in subsequent processes.

Next, as described above, a process of classifying compactly represented feature map sets, that is, compact representations, is performed. This process may be performed by the template comparison block 15, as follows.

First, the template comparison block 15 calculates a distance between a representation object to be classified and a predetermined registered image registered in the registered image database 14, that is, the distance between the templates. In other words, a similarity measurement is performed between a template, which is a registered image, and a template, which is a compact representation of an input image to be tested.

Then, the grade of the input image to be tested is determined based on the calculated distance, that is, the similarity measurement result. That is, based on the similarity calculation, it is determined which registered image having the most similar among the registered images registered in the registered image database 14, and determines the grade of the determined registered image as the grade of the input image to be tested. (S170).

In an embodiment of the present invention, an existing video observation system is automated according to the person identification method as described above, and the method is used for the production of a new class of intelligent biometric person recognition system according to a special list. In this case, the devices constituting the system are implemented in the form of a plate, and the elements for performing the method as shown in FIG. 2 must be assembled thereon, and at least one signal processor in order to perform necessary operations. ) May be included.

A suitable method for applying the system according to an embodiment of the present invention is to increase the effectiveness of the existing video observation system, and to produce a new class of intelligent system in consideration of additional types of types such as intelligent video observation and special recognition according to a special list. It is to be used to. The instrument is then implemented in the form of an equipment plate containing at least one signal processor.

The embodiments of the present invention described above are not only implemented through the apparatus and the method, but may also be implemented through a program for realizing a function corresponding to the configuration of the embodiment of the present invention or a recording medium on which the program is recorded. Implementation may be easily implemented by those skilled in the art from the description of the above-described embodiments.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

1 is a structural diagram of a person identification system according to an embodiment of the present invention.

2 is a flowchart illustrating a person identification method according to an embodiment of the present invention.

3 is an exemplary view illustrating a feature point map manufacturing process according to an exemplary embodiment of the present invention.

Claims (10)

In the method of identifying a person, Forming a feature point map of an image to be tested; Measuring a similarity between a feature point map of the image and registered images already processed and registered in a database; And Determining a grade of the image under test based on the result of the similarity measurement Including, person identification method. The method of claim 1 The determining of the similarity may include determining whether the two images are identical by using an image feature point map set classification algorithm that performs two sets of similarity measurement calculations. The method of claim 1 Forming the feature point map, And the feature point map is formed based on a method of convolving a gradual directional filter core and an image matrix. The method of claim 3, The stepwise directional filter core is regarded as a complex core and forms the feature point map using absolute values of the convolution result. The method of claim 3, And treating the stepped directivity filter core as a composite core and forming the feature point map using the phases of the convolution results. The method of claim 1 Filtering the feature point map and forming a feature point map having an even distribution based on a method of all or partial arrangement of the filtering result. The method of claim 1, And converting a matrix corresponding to the feature point map of the image into a matrix having a distribution of values already defined. A face detection block for detecting a face from an image of a person obtained by a camera; An image preprocessing block which performs preprocessing on the detected face image; A coding block for creating a feature point map for the preprocessed image and processing the feature point map into a set compact representation; A registration image database storing registration images for the subjects; And A template comparison block that measures the similarity between the registered image and the image according to the obtained expression and determines the grade of the image based on the similarity. Including a person identification system. The method of claim 8, The registered image database includes data for comparing templates and making decisions, Wherein said coding block comprises a tool for producing a feature point map of an image and a tool for converting values of a matrix corresponding to the feature point map of an image into a matrix having a distribution of predetermined values. The method of claim 8, The template comparison block includes a basic image classifier that is implemented to measure similarity to two images and to determine recognition based on the results.

Images