KR101105435B1 - Face detection and recognition method - Google Patents

Abstract

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to face detection and progressive face recognition through progressive visual environment recognition using biological based salience maps using image processing technology. We use the initial visual information such as brightness, color, contour, and symmetry axis information to obtain the protrusion map that determines the prominent area more than the surrounding area. Then we detect the face through visual environment recognition. Provides a progressive face recognition method. To this end, the present invention is a pre-processing step of recognizing the environmental changes in the input image through the visual environment recognition; A detection step of extracting a protrusion map based on the environment change obtained in the preprocessing step and detecting a face using a face detection algorithm in the protrusion map; Disclosed is a method comprising a face recognition through a face recognition algorithm and a face recognition step of progressively recognizing an unlearned face by discriminating a face that has not been learned and a face that has been learned. According to the present invention, a face that has not been previously learned among the detected faces may be gradually recognized and accurately recognized.

Protrusion map, IHMAX, IPCA, progressive face recognition, face detection, environment recognition

Description

Face detection and recognition method

The present invention relates to face detection and gradual face recognition using gradual visual environment recognition using a biological-based protrusion map using image processing technology. More specifically, by modeling the human visual mechanism, by using the initial visual information such as brightness, color, contour, and symmetry axis information for the input image, it obtains the protrusion map that determines the prominent area more than the surroundings. Face detection and progressive face recognition methods.

Currently, the face recognition method is used in various fields such as a robot device, security surveillance, and a mobile communication terminal. There are various methods of detecting and recognizing face regions in a video including a human face or a video such as a still image or an image.

A detector and a detection method (Korean Patent Application No .: 1020060055959) for detecting a face candidate region by using a face color from an input image and detecting a face by using an adaboost for the detected face candidate region, and a complex pattern of a face. Method of detecting face region using Principle Component Analysis (PCA) and Linear Discriminant Analysis (LDA) that can be represented by several principal component values (Korean Patent Application No. 1020020070810) A method of pre-learning a SVM (Support Vector Machine) with excellent recognition performance without using a face and a non-face and verifying a face using a learned database (Korean Patent Application No. 1020050070751), and a two-dimensional Gabor filter And a method of recognizing a face by using the same (Korean Patent Application No. 1019980027314).

However, these prior arts do not consider human visual processing mechanisms at all and do not provide a gradual face recognition method. In other words, research and development of self-identification and security system through face recognition at home and abroad is gradually increasing its market size and positioning itself as a promising industry, but currently, self-identification and security system performs face recognition based on stored face data. It is very difficult to learn new face data.

The present invention models a human visual mechanism and obtains a projection map that determines the prominent area more prominently than the surrounding area by using initial visual information such as brightness, color, contour, and symmetry axis information on the input image, and then detects the face through visual environment recognition. For the purpose.

In addition, unlike the prior art in which face recognition for a person who has not been learned in the past provides a progressive face recognition method.

In addition, the present invention improves the efficiency in face detection through the face detection step according to the progressive visual environment in the input image and detects the face in the protrusion map, thereby increasing the detection speed and thereby gradually learning and recognizing the detected face. It aims to provide a self-identification and security system with high adaptability and high recognition performance.

In order to achieve the above object, the present invention provides a method for face detection and progressive face recognition using gradual visual environment recognition using a biological-based protrusion map, in a limited condition and environment not solved by conventional image processing technology. Generating a protrusion map using an existing image processing technology in order to be robust to an environment using a protrusion map created by imitating a human visual function, and to a more general environment; Recognizing a visual environment by using a distance of each protruding point of an input image and an energy value of the protruding region using the protruding map; Thereafter, a face candidate region is determined by generating a protrusion map for detecting a face region of a person, and preprocessing of face recognition for determining whether the determined face candidate region is a face; And extracting information required for progressive face recognition using a feature point / feature vector extraction algorithm to provide a face detection and a progressive face recognition method through a biological-based progressive environment recognition that mimics the human visual function.

For another object of the present invention, the region of interest separation step of separating / extracting only the region (hereinafter referred to as the region of interest) that contains the element of interest generating the protrusion map using the image processing technology; Determining a change of an existing input image through gradual environment recognition using a protrusion map; It provides a face detection and gradual face recognition method through gradual environment recognition based on biology, including information extraction step for progressive object recognition using feature point / feature vector extraction algorithm.

In addition, the present invention provides a method for recognizing an environment from an image using an image processing technique, the method comprising: a preprocessing step of obtaining an information that is not affected by the surrounding environment from an input image; An object-oriented protrusion map forming step of determining an object candidate by obtaining an object-oriented protrusion map from the image obtained in the preprocessing step; An environmental cognitive change checking step of detecting a person's appearance in the image by changing the image through the environmental cognition; A face oriented protrusion map forming step of forming a face oriented protrusion map to find a face candidate region of a person in an image; A face detection step within a specific area for detecting a complete face area from the face candidate area; A feature-based face recognition step of extracting a feature from the detected face and recognizing a face based on the feature; And, in the case of the face that has not been learned, face detection and gradual face recognition method using gradual visual environment recognition using gradual face feature extraction step including progressive face feature extraction step of extracting gradual face feature for further learning is disclosed. do.

The object-oriented protrusion map forming step may include: (a) extracting color information and shape information from an image; (b) extracting color and shape features from each piece of information; (c) combining the features to create an object-oriented protrusion map; And (d) extracting the object candidate regions in the order of being more prominent than the surroundings in the object-oriented protrusion map.

The environmental cognitive change check step may include: (e) extracting distance information between protrusions; (f) combining the distance information and the energy information between the protruding points; And (g) observing the combined information from step (f) above.

The forming of the face-oriented protrusion map may include: (h) extracting color information and shape information from an image; (i) determining a range of human skin color from color information and separating only an image of a region of interest; (j) extracting color features or shape features from information of the ROI; (l) combining the features to create a face oriented protrusion map; And (m) extracting the face candidate regions in a more prominent order than the surroundings in the face-oriented protrusion map.

The face detection in the specific region may be performed by detecting an actual face region by analyzing an image difference in an image by applying an AdaBoost algorithm based on a face candidate region.

A feature may be extracted from the detected real face area.

The face may be recognized based on the extracted features.

If the face recognition result is a learned face after the feature-based face recognition step, reflecting corresponding information; And gradually extracting facial features to add data of the person to the face learning data when the face recognition result is not the learned face.

And re-learning face recognition with the progressive face feature information.

Extracting color information and shape information from the image may include extracting red, green, and blue information and contrast information as color information; And extracting contour information from the shape information, and extracting the feature information includes obtaining a pyramid image of each of the information images and acquiring the feature of each of the information images through a center-peripheral image difference. Can be.

Extracting a relative distance between the points with distance information between the protrusion points obtained from the image; Extracting an energy value of each protrusion area from the protrusion degree information of each protrusion point in the image; The method may include combining the distance information between the protrusion points and the energy value of the protrusion area.

The method of determining a range of human skin color from the color information and separating only the corresponding area comprises: finding a range of human skin color; Obtaining maximum and minimum values of the red, green, and blue images in the corresponding range; And separating the image into the obtained boundary values.

If the face recognition result is not a learned face, the method may include a step of including information of the corresponding person in a database for real-time learning while gradually extracting features.

In this way, the present invention models the human visual mechanism and obtains a projection map that determines the prominent area more prominently than the surrounding area by using initial visual information such as brightness, color, contour, and symmetry axis information on the input image. By performing face detection through, face detection performance is very excellent.

In addition, the present invention can gradually recognize and recognize the face, unlike the prior art that does not recognize the face for a person who has not been previously learned.

In addition, the present invention improves the efficiency in face detection through the face detection step by gradually recognizing the progressive visual environment in the input image and detects the face in the protrusion map, thereby increasing the detection speed and thereby gradually learning and recognizing the detected face. This provides a highly adaptive and highly aware identity authentication and security system.

In addition, the present invention will be able to develop a unique product with the technology developed as described above as a solution of the inconvenience factor of many identity authentication of life, or can lead to the performance improvement and automation of the security system. Furthermore, the progressive face detection and face recognition technology through progressive visual environment recognition according to the present invention can be applied to various fields such as identity authentication and security system, industrial field and military use by utilizing not only face but also object recognition.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Here, in the case where it is obvious to those skilled in the art that the detailed description of the known configuration or function may obscure the subject matter of the present invention, the detailed description thereof will be omitted.

1 is a diagram illustrating a schematic configuration of a face detection and a gradual face recognition method through gradual visual environment recognition using a biological-based protrusion map according to the present invention.

As shown in FIG. 1, face detection and progressive face recognition using progressive biological environment recognition using a biological-based protrusion map according to an embodiment of the present invention include a photographing step (S1), a preprocessing step (S2), and an object orientation. Protrusion map forming step (S3), environmental cognitive change checking step (S4), face-oriented protrusion map forming step (S5), face detection in a specific area (S6), detected face feature extraction step (S7), feature-based face A recognition step S8, a learned face presence determination step S9, a face recognition step S10, and a gradual face feature extraction step S11 are included.

In the photographing step S1, the photographed image photographed in the current environment using the camera is output to the preprocessor at predetermined time intervals.

In the preprocessing step (S2), the image output from the camera is calibrated and corrected using the preprocessor to perform image processing.

In the object-oriented protrusion map forming step (S3), the protrusion map is formed by using the result output from the preprocessor using the object-oriented protrusion map forming unit and the existing image processing technology.

More specifically, the object-oriented protrusion map forming step S3 may include extracting color information and shape information from an image, extracting a color feature and a shape feature from each piece of information, and combining the respective features. The method may include creating an oriented protrusion map and extracting object candidate regions in the order of being more prominent than the surroundings in the object oriented protrusion map.

Here, the extracting of color information and shape information from the image may include extracting red, green, and blue information and contrast information from color information, and extracting contour information from shape information.

The extracting of the color feature and the shape feature is performed by obtaining a pyramid image of each information image and acquiring the feature of each information image through a center-peripheral image difference.

In the environmental cognitive change check step (S4), a change in the image is performed through environmental cognition to check whether a person appears in the image. That is, the ROI is extracted based on the protrusion map configured as described above. This is to determine whether the region of interest created from the input image from the current camera is different from the region of interest extracted from the previous frame in order to recognize the progressive environment.

Here, since the protrusion map that mimics the human visual function is effective for extracting the region of interest, the embodiment of the present invention implements progressive environment recognition in the object-oriented protrusion map forming unit based on the existing image processing technology.

Biological background The hippocampus organ of the brain is known to remember input video scenes and is responsible for generating topology of search paths that provide more information.

Therefore, the present invention assumes that the scene can be stored based on the topology of the protruding regions obtained from the obtained protruding map information. In addition, all the protruding regions can be sized using entropy maximizing method, and the protruding regions of the input image can be represented in a topology. In addition, the energy information obtained from the protrusion map is an additional factor in determining the presence or absence of a change in the scene. The acquired protrusion topology and energy information are used as input to the model, and the model gradually stores these two inputs.

The present invention can recognize the direction and the location of the gaze from the stored scene information, and implemented a habituation algorithm by detecting the degree of repetition of the image.

As an example, the environmental cognitive change check step (S4), the step (e) of extracting the distance information between the protruding points, the step (f) of combining the distance information and energy information between the protruding points, and (f) Observing the combined information from step (g).

In other words, the environmental cognitive change check step (S4) is a step of extracting the relative distance between each point with the distance information between each protrusion point obtained from the image, and the protrusion degree information of each protrusion point in the image The method may include extracting an energy value of each protruding region, combining distance information between the protruding points, and an energy value of the protruding region.

In the face-directed protrusion map forming step (S5), a face-oriented protrusion map is formed to find a face candidate region of a person in the image.

For example, the step of forming the face-oriented protrusion map (S5) may include extracting color information and shape information from an image (h), determining a range of human skin color from the color information, and separating only an image of the ROI (i). ), Extracting color and shape features from the information of the region of interest (j), combining each of the features to create a face-oriented protrusion map (l), and more prominent than the surroundings in the face-oriented protrusion map. Extracting a face candidate region may be performed (m).

Here, the method of determining the range of human skin color from the color information and separating only the corresponding area may include finding a range of human skin color, obtaining maximum and minimum values of red, green, and blue images in the corresponding range, and obtained boundary values. The image may be separated into a step.

In the specific area face detection step S6, a complete face area is detected from the face candidate area. The face detection step S6 in the specific region may be performed by detecting an actual face region by analyzing an image difference in an image by applying an AdaBoost algorithm based on a face candidate region.

In the detected face feature extraction step S7, a feature is extracted from the detected real face area.

In the feature-based face recognition step S8, a face is recognized based on the extracted features.

In the learning face determination step (S9), it is determined whether the recognized face is already a learned face.

If the face has already been learned in step S9, the face recognition step S10 is performed. Otherwise, the progressive face feature extraction step S11 is performed.

First, in the face recognizing step (S10), the face is recognized by reflecting the corresponding information.

Next, in the progressive face feature extraction step (S11), a gradual face feature is extracted so that the person can add data to the face learning data. That is, in the progressive face feature extraction step S11, face recognition is relearned with the progressive face feature information. In other words, in the progressive facial feature extraction step (S11), the information of the person is included in the database for real-time learning, but the progressive feature extraction is performed.

2 is for explaining a gradual visual environment cognitive process using a biological-based protrusion map applied to the present invention.

As shown in FIG. 2, the present invention performs progressive visual environment recognition by combining a scene change detection model with a visual attention model. That is, the bottom-up attention model and the top-down attention model are combined to generate the protrusion map, and then the scene change detection model indicates whether there is a change of the scene in the input image from the protrusion map information.

3 is for explaining a scene change detection model applied to the present invention.

As shown in FIG. 3, scene change detection is performed in the space-time dimension to perform scene memory. First, a scene change is detected by comparing a stored scene with a newly received input image. The information used at this time is the similarity between the stored scene information and the new input image. After measuring the similarity, the similarity creates a nodule for a new scene change. Thus, by comparing the stored nodules with the newly generated nodules, the positional information can be obtained through the most similar nodule information.

If the input video is already viewed and cannot detect a change in the scene, the input video is displayed. Thus, when it is determined that the input image has been viewed previously, it is possible to express habituation.

4 is for explaining the results of a gradual visual environment cognition experiment using a biological-based protrusion map applied to the present invention.

As shown in FIG. 4, the three images in the upper row all view the same image. However, all three images have slightly different fields of view. Although the first image and the second image have slightly different clocks, but there is no change in the image, it can be seen that the topology of the gaze path based on the protrusion map is similar. The third image is a case where there is a change. In this case, the gaze path topology is completely different from the gaze path topology for the two previous images. That is, it can be seen that the novelty of the scene can be detected using the topology of the gaze path obtained from the protrusion map of the input image. The three images in the lower row are projection map images for each image in the upper row.

Here, referring again to FIG. 1, in the object-oriented protrusion map, the distance and the salvage area of each of the salient points of the input image are separated using the biologically-based protrusion map with the image processed by the preprocessor. Progressive visual environment recognition is performed using the energy value of.

When the input image from the camera is changed through the gradual recognition of the visual environment, face detection in a specific area is performed through a biological-based face-oriented protrusion map forming unit.

5 is for explaining a face detection model in a specific region using a biological-based face-oriented protrusion map applied to the present invention.

As shown in FIG. 5, the face region detection algorithm considers biological aspects in a way that can overcome the limitations of a computer vision system.

In a real environment, an input image is often complicated by a complicated background or a change of environment. AdaBoost doesn't perform well in this environment because it only takes shape information into account. So, in order to overcome these limitations, the area of interest (face area) is obtained through a biological-based protrusion map using a color filter, and to determine whether there is a face or another object in this area, AdaBoost ) To detect the face.

In addition, the protrusion map formed by the feature map for extracting the size information of the ROI is a brightness value that maximizes the variation value in the histogram using Otsu's threshold technique. The bin area is binarized, and the labeling operation is used to obtain the size information of the bin area.

This not only improves the computational speed of the real-time system, but also increases the size of the object depending on the degree of protrusion of the object, regardless of the distance. The size of the area can be determined.

In addition, after determining the size of the protruding region through the above operation process, to determine whether the object overlaps each other, the histogram distribution in the X-axis and Y-axis directions in the binary image of each labeled object is checked. When the objects overlap with each other, the overlapping objects are separated by using the bending point information of the histogram.

FIG. 6 illustrates a process of separating a face region using histogram information in a face-oriented protrusion map applied to the present invention.

As shown in FIG. 6, in the present invention, the face area is separated by using histogram information in the protrusion map.

After the aforementioned process, the final face area is detected and verified using AdaBoost. AdaBoost is a state-of-art technology of face detection algorithms, which is a learning algorithm that detects a face using shape information using a Haar-like wavelet. The algorithm's internals consist of a combination of simple linear classifiers that are not very good at classifying, allowing for more accurate classification. Trying to use this learning algorithm requires positive data and negative data. Thus, attempting to use it for face detection requires face image data that is positive data and data that is not face that is negative data.

FIG. 7 illustrates extraction of shape information using a haar-like wavelet for detecting and verifying a facial region applied to the present invention.

As shown in FIG. 7, the positive data and the negative data are extracted from the shape information through a haar-like wavelet, and the feature information is extracted from an adaboost ( AdaBoost) is used as learning data.

FIG. 8 illustrates results of a face detection experiment in a specific region using a biological-based face-oriented protrusion map applied to the present invention.

As shown in FIG. 8, when an input image is input, a color filter is first applied to the intensity, edge, and RG complementary color information and applies a color range produced by experiments several times. To filter the video.

Then, three feature information is extracted through center-surround difference and normalization, and a salient map is generated from this. The ROI is obtained by using the information of the saliency map. This becomes the face candidate images used later for face recognition.

Finally, the face candidate images are determined whether or not the face is using AdaBoost.

In addition, the face detection algorithm applied to the present invention mimics the initial visual mechanism of a person, and may detect a face region candidate in real time and determine whether it is a face using AdaBoost. In natural images with complex backgrounds and diverse environments, the algorithm not only detects non-face areas but also recognizes faces. The algorithm can further improve the performance of AdaBoost using a biologically based attention visual function. This approach has shown that biologically based approaches in vision system development can overcome existing limitations or limitations.

Referring back to FIG. 1, as a next step for face detection and gradual face recognition using gradual visual environment recognition using a biological-based protrusion map, a gradual face feature information extraction process applied to the present invention is applied to the present invention. Models include progressive feature extraction model using incremental principal component analysis (ICPA) and progressive feature extraction model using incremental hierarchical MAX (IHMAX).

First, the progressive feature extraction model using the IPCA used an IPCA introduced by Hall and Martin to gradually extract the features of the existing face.

The algorithm of IPCA is as follows. The initial eigenvectors and eigenvalues are obtained from covariances with the first gathered N training data, and k principal components are selected to represent the training data through projective operation. When a new data (N + 1) is received, IPCA is applied to form a new eigenvector and eigenvalue from the current eigenvector U _{m * k} and eigenvalues (m is the dimension of the entire image, k is Selected Dimension) Equation 1 is an equation for updating the mean for a new input y.

A new vector is added to the new input data y and the rotation vector R is applied to form a new eigenvector. Equation 2 is a method of obtaining a new eigenvector by rotating eigenvectors in an added vector space using a rotation vector R in a vector space where a new vector is added in a direction perpendicular to the current vector.

이고,

로 벡터

는 새로 더해지는 벡터로서 다른 고유벡터에 직교한다. 새로운 고유벡터를 갱신하기 위해 (k+1)(k+1)차원의 수학식 3, 4로 생성되는 공분산행렬을 인터메디에이트 아이겐프러블럼(intermediate eigenproblem) 방법을 이용하여 풀어야 한다. here

ego,

Vector

Is a newly added vector that is orthogonal to another eigenvector. In order to update a new eigenvector, the covariance matrix generated by the equations (3) and (4) in the (k + 1) (k + 1) dimension must be solved using the intermediate eigenproblem method.

,

이다. 적절한 특징들의 차원을 결정하기 위해 새로운 벡터가 추가될지를 결정하는 함수 A(k)는 수학식 5에 의해 결정된다.here

,

to be. The function A (k) that determines whether a new vector will be added to determine the dimensions of the appropriate features is determined by equation (5).

If the A (k) value calculated above is smaller than the specific threshold value, the new eigenvector is added to the existing eigenvector. If the A (k) value is greater than the specific threshold value, only the existing eigenvector is obtained. The threshold value indicates how efficiently the dimension of the data can be represented according to the experiment of the image data.

That is, when a new face is input, if the newly input face belongs to the existing class, the eigenvector is not changed and the data input as the original eigenvector is represented. However, if the newly input data does not belong to the existing class, the eigenvector and eigenvalue are added to the current vector, and the input data is obtained in a gradual way. Color information among the elements used as input of IPCA is used to convert red, green, and blue RGB colors extracted from the input image into HIS (hue, intensity, saturation) colors. Doing. However, due to the constantly changing projected area size, sampling is required to compensate for the change in the size of each input information and convert it into input information of the same size. Therefore, we can input robust feature patterns of IPCA for various size changes using a log-polar sampling method that is similar to human visual mechanisms and robust to size and rotation conversion.

9 illustrates a log-radial sampling process for IPCA input data having a fixed size for progressive facial feature extraction applied to the present invention.

As shown in FIG. 9, the input pattern of the feature map for the protruding region having various sizes may be converted into the input size of the fixed IPCA using log-radial sampling.

The feature dimension of the input color for expressing the selected face feature is determined by the number of eigenvectors selected from each of the independent features of hue, saturation, and contrast. An initial shape feature information is extracted using an 8-direction Gabor filter to express shape information of an input face. The input determined by the winner-take-all method of the contour information of the contrast, R + G-, and B + Y- color feature maps in order to reflect the superior feature information in the prominent area as the input of the Gabor filter. use. The IPCA is applied to reduce the dimension of feature information in eight directions and to extract important feature information. The input dimension of the shape feature information is determined by the number of eigenvectors obtained from the IPCA in eight directions. Finally, the object can be represented in a reduced dimension determined by the number of eigenvectors of color and shape feature information.

FIG. 10 illustrates a facial feature expression model using IPCA for gradual facial feature extraction applied to the present invention.

As shown in FIG. 10, it can be seen that gradual face feature extraction is possible using a gradual method.

Table 1 below shows recognition performance results in an environment in which feature points to be recognized using IPCA are gradually increased. The experiment represented data by obtaining eigenvectors through PCA with data belonging to three of 23 classes. When 75 new data is used as an input and belongs to an existing class, it is represented as the same class, and when it does not belong to a new class, it is judged as correct (correct) to show performance.

Right
(correct) error
(wrong) Recognition rate
(recognition rate)
(75) 54
(72%) 21
(28%)

As shown in the experiment results, when new object data that has not been previously learned belongs to the class used for learning, new eigenvectors are not added in IPCA, and the object is expressed using the eigenvector like the existing PCA. For new object data that does not belong, a new eigenvector is added by IPCA, a new class is created, and each object is localized. In other words, IPCA implements incremental knowledge accumulation and additional learning functions about human objects.

Next, the gradual feature extraction model using incremental hierarchical MAX (IHMAX) applied to the present invention can be seen as a more effective system made by mimicking the human brain mechanism.

The conventional hierarchical maximum feature extraction (HMAX) model is also a model that mimics this brain mechanism.

However, this model recognizes new objects using the 2000 criteria already learned, and therefore does not perform well when features that differ from the previously learned criteria are introduced. And there are disadvantages in speed by always using 2000 criteria.

Thus, in the model applied to the present invention, it is possible to recognize the object more effectively by recognizing the feature point using criteria gradually increasing according to the environment, not 2000 criteria applicable only in such limited environment.

Applied to the present invention, the model is largely divided into a feature extraction method and a classifier capable of classifying variable features. In feature extraction, up to C1 of the HMAX model, two layers are used to express the spatial component by selecting a spatially larger value. To make the feature more robust, it is represented by a histogram of the directional components and put into the GCS to create a standard suitable for a new environment that eliminates noise and peak components. This criterion increases or stays current depending on the image being heard. Using this increasing standard, feature information is extracted by comparing with the input image. Since the extracted information increases in dimension according to the criteria, it cannot be distinguished by a general classifier.

The model used in the model applied to the present invention is a Gaussian mixture model of a hierarchical structure with a variation of the generic model.

This classifier model is used to process progressively increasing feature information, allowing for progressive representation and recognition of feature points.

FIG. 11 illustrates a gradual feature extraction model using incremental hierarchical MAX (IHMAX) for gradual facial feature extraction applied to the present invention.

As shown in FIG. 11, after a feature extraction process of a gradual face is performed, a face is recognized if it is a learned face after discriminating whether it is a previously learned face feature or a new face feature. Face recognition.

Face detection and gradual face recognition using gradual environment recognition using a biological-based protrusion map applied to the present invention has been described as an example for face detection and gradual face recognition in the above-described embodiments, but is not limited thereto. . In other words, it is possible to use non-face objects, not floating, and can be used in various places. It is possible to show better performance in future developed system.

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 not intended to limit the scope of the present invention but to limit the scope of the technical idea of the present invention. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

1 is a diagram illustrating a schematic configuration of a face detection and a gradual face recognition method through gradual visual environment recognition using a biology-based salient map according to the present invention;

2 is a view for explaining a gradual visual environment recognition process using a biological-based protrusion map applied to the present invention,

3 is a view for explaining a scene change detection model applied to the present invention,

4 is a view for explaining the results of a gradual visual environment cognition experiment using a biological-based protrusion map applied to the present invention,

5 is a view for explaining a face detection model in a specific region using a biological-based face-oriented protrusion map applied to the present invention;

6 is a view for explaining a separation process of a face region using histogram information in a face-oriented protrusion map applied to the present invention;

FIG. 7 is a view for explaining shape information extraction using a Haar-like wavelet for detecting and verifying a facial region applied to the present invention; FIG.

8 is a view for explaining the results of the face detection experiment in a specific region using a biological-based face-oriented protrusion map applied to the present invention,

9 is a view for explaining a log-radial sampling process for IPCA input data having a fixed size for progressive facial feature extraction applied to the present invention;

10 is a view for explaining a facial feature expression model using IPCA for gradual facial feature extraction applied to the present invention,

11 is a view for explaining a gradual feature extraction model using incremental hierarchical MAX (IHMAX) for gradual facial feature extraction applied to the present invention.

Claims (13)

A preprocessing step of calibrating and correcting an image input by a camera at predetermined time intervals; An object-oriented protrusion map forming step of determining an object candidate by obtaining an object-oriented protrusion map from the image obtained in the preprocessing step; Determining whether an image input from the camera is different from a previously input image, and detecting an appearance of a person; A face oriented protrusion map forming step of forming a face oriented protrusion map to find a face candidate region of a person in an image; A face detection step within a specific area for detecting a complete face area from the face candidate area; And, And a feature-based face recognition step of extracting color and shape features from the detected face and recognizing a face based thereon. The object-oriented protrusion map forming step, (a) extracting color information and shape information from an image; (b) extracting color and shape features from each piece of information; (c) combining the features to create an object-oriented protrusion map; And, (d) extracting object candidate regions in the order of prominence from the surroundings in the object-oriented protrusion map; The environmental cognitive change check step, (e) extracting distance information between protrusions; (f) combining energy values of each protrusion area with distance information between protrusion points and protrusion degree information of the protrusion points; And, (g) observing the combined information from step (f) above; The face-oriented protrusion map forming step (h) extracting color information and shape information from the image; (i) determining a range of human skin color from color information and separating only an image of a region of interest; (j) extracting color and shape features from information of the region of interest; (l) combining the features to create a face oriented protrusion map; And, (m) extracting face candidate regions in order of being more prominent than the periphery in the face-oriented protrusion map; After the feature-based face recognition step If the face recognition result is a pre-learned face, reflecting a corresponding color feature and a shape feature; And, And extracting a face feature to add data of the person to face learning data when the face recognition result is not a pre-learned face. And re-learning face recognition with the face feature information. delete delete delete The method of claim 1, The face detection step in the specific region A face detection and face recognition method comprising detecting real face areas by analyzing image differences in an image by applying an AdaBoost algorithm based on face candidate areas. The method of claim 5, Extracting color and shape features from the detected real face region. delete delete delete The method of claim 1, During the object-oriented protrusion map forming step Extracting color information and shape information from the image, Extracting red, green and blue information and contrast information as color information; And Extracting contour information as shape information; Including, During the object-oriented protrusion map forming step Extracting the feature information, Obtaining a pyramid image of each of the information images and acquiring features of the respective information images through a center-peripheral image difference; Face detection and face recognition method comprising a. delete The method of claim 1, During the face-oriented protrusion map forming step Determining the range of human skin color from the color information and separating only the image of the ROI Finding a range of human skin colors; Obtaining maximum and minimum values of the red, green, and blue images in the corresponding range; And, Separating the image into boundary values which are maximum and minimum values of the obtained red, green, and blue images; Face detection and face recognition method comprising a. delete

Images