KR20170096924A - Collaborative facial color feature learning method of multiple color spaces for face recognition and apparatus therefor - Google Patents

Abstract

A cooperative facial color feature learning method in various color spaces and an apparatus therefor are disclosed. The cooperative facial color feature learning method according to an embodiment of the present invention includes a step of extracting a color feature from each of a plurality of predetermined color spaces for a facial image; and a step of generating a fused feature with the extracted color features based on the learning of the extracted color features. The step of extracting color features can extract the color feature by learning the feature of deep convolutional neural networks (DCNNs) of each of the color spaces. Face recognition rate can be improved.

Description

Field of the Invention < RTI ID = 0.0 > [0001] < / RTI > A method and apparatus for cooperative facial color features in various color spaces,

The present invention relates to cooperative facial color feature learning, and more particularly, to a method and apparatus for enhancing facial recognition rate using fused features by fusing extracted color features in various color spaces through learning will be.

Facial color is an important factor in face recognition, and different color spaces provide complementary information. Especially, face recognition accuracy using face color information is known to provide improved recognition rate in illumination change and low resolution environment.

Many face color based face recognition frameworks have been studied for decades, and the technique of one embodiment develops a monochrome face recognition framework as a color face recognition framework, and the technique of another embodiment maximizes the complementary effect , Local color vector binary patterns (LCVBPs) have been proposed to represent distinctive texture patterns obtained from spatial interactions between different spectral band images.

To obtain a more discernible color space in face recognition, a color image identification (CID) space based on a face recognition framework was designed. In the R, YIQ color space of RGB color space and RQCr using Cr of YCbCr color space A named hybrid color space configuration has also been proposed.

However, the accuracy of facial recognition according to the selected color space may vary depending on the case of the face recognition application and the environment.

 In order to obtain the best facial recognition rate when selecting a color space from various color spaces, a framework for selecting color-element features has been proposed. The deep learning framework has attracted much attention among researchers because of its high image recognition rate, When the running framework is applied to face recognition, the face recognition rate is significantly increased.

Currently, RGB color images (or black and white images) are usually used in facial recognition deep learning, and research on color utilization in the deep learning framework has not been well under way. Despite the well-known advantages of facial color and its utility, the effect of facial color in deep running is still an incomplete problem.

Therefore, there is a need for a method for improving the face recognition rate.

Embodiments of the present invention provide a method and apparatus for enhancing a face recognition rate using fused features by fusing extracted color features in various color spaces through learning.

A face color feature learning method according to an exemplary embodiment of the present invention includes extracting a color feature from each of a plurality of predetermined color spaces for a face image; And generating the fused feature with the extracted color features based on learning of the extracted color features.

The extracting of color features may extract the color features by learning features of deep convolutional neural networks (DCNNs) of each of the color spaces.

The generating the convergence feature may generate the convergence feature by concatenating all of the feature vectors for the extracted color features.

The generating the fusion feature may generate the fusion feature by learning deep neural networks (DNNs) for the extracted color features.

The DNN may comprise a plurality of fully connected layers.

According to another aspect of the present invention, there is provided a face recognition method using a facial color feature, the method comprising: extracting a color feature from each of a plurality of predetermined color spaces for a face image; Generating a fusion feature in which the extracted color features are fused based on learning of the extracted color features; And recognizing a face of the face image based on the generated fusion feature.

The extracting of color features may extract the color features by learning features of deep convolutional neural networks (DCNNs) of each of the color spaces.

The step of generating the convergence feature may generate the convergence feature by learning deep neural networks (DNN) composed of a plurality of fully connected layers for the extracted color features.

According to an aspect of the present invention, there is provided an apparatus for learning facial color features, comprising: an extractor for extracting color features from each of a plurality of predetermined color spaces with respect to a face image; And a fusion unit for generating a fusion feature in which the extracted color features are fused based on learning of the extracted color features.

The extraction unit can extract the color feature by learning features of deep convolutional neural networks (DCNN) of each of the color spaces.

The fusion unit may generate the fusion feature by concatenating all feature vectors for the extracted color features.

The fusion unit can generate the fusion feature by learning deep neural networks (DNN) for the extracted color features.

The DNN may comprise a plurality of fully connected layers.

According to an aspect of the present invention, there is provided an apparatus for recognizing a face using a facial color feature, the apparatus comprising: an extractor for extracting a color feature from each of a plurality of predetermined color spaces with respect to a face image; A fusion unit for generating a fusion feature in which the extracted color features are fused based on learning of the extracted color features; And a recognition unit for recognizing a face of the face image based on the generated fusion characteristic.

The extraction unit can extract the color feature by learning features of deep convolutional neural networks (DCNN) of each of the color spaces.

The fusion unit can generate the fusion feature by learning DNNs (deep neural networks) composed of a plurality of fully connected layers for the extracted color features.

According to embodiments of the present invention, the extracted color features in various color spaces are fused through learning, thereby improving the face recognition rate using the fused features.

Figure 1 shows an exemplary diagram of a DCNN structure as a reference in color space.
Figure 2 shows examples of learned filter kernels in different spaces.
FIG. 3 is a diagram illustrating an example of a cooperative facial color feature learning method according to an embodiment of the present invention.
Fig. 4 shows a feature map in the first convolution layer C1 for each color space of the face image.
5 shows a feature map in the second convolution layer C2 for each color space of the face image.
Figure 6 shows an exemplary view of a color space and a two-dimensional feature subspace for a test image.
FIG. 7 illustrates a configuration of a cooperative facial color feature learning apparatus according to an embodiment of the present invention.
FIG. 8 shows a configuration of a face recognition apparatus using a face color feature according to an embodiment of the present invention.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to or limited by the embodiments. In addition, the same reference numerals shown in the drawings denote the same members.

Embodiments of the present invention are intended to improve the face recognition rate using fused features by fusing extracted color features in various color spaces through learning.

FIG. 1 shows an example of a reference DCNN structure in a color space. As shown in FIG. 1, a reference DCNN includes four convolution layers C1, C2, C3 and C4 ), Two subsampling layers (or Max Pooling layers S1 and S2) and two fully connected layers (L = 8).

Here, the fully connected layer can consist of an output layer consisting of F1 and Softmax functions

The input face image size is obtained by convolution with 32x32 pixels and 32 feature maps for 3 color elements with 32 kernels (represented by 32 @ 3x3) of 3x3 size, The loop operation is performed as shown in Equation (1) below.

[Equation 1]

Here, h _j ^(l) is l means the j-th feature map in the second layer, and n _l means a number of characteristic maps in the l-th layer, and w _ij ^(l) refers to the weight, and b _j ^{(l ) May} mean bias.

DCNN is learned by weight and bias, and the first convolution layer of the color face image can be calculated as Equation (2) below.

&Quot; (2) "

Where C _k = {C _k ¹ , C _k ² , C _k ³ } and can mean three color elements. For example, in the case of an RGB image, C _k ¹ may denote R, C _k ² may denote G, and C _k ³ may denote B. That is, in the first convolution layer that receives the color facial image as an input, the convolution may be a weighted sum of color elements having DCNN weighting parameters.

The weights learned in the DCNN can have different shapes depending on the color space of the input face image.

Fig. 2 shows examples of learned filter kernels in different spaces. In Fig. 2, a first convolve learned in color spaces of RGB (a), La * b * (b), YCbCr (c) and HSV Here is an example of 3x3 filter kernels (DCNN weights) in the routing layer.

As shown in FIG. 2, the weights have different shapes, and the activation maps corresponding to the respective input color face images have different response results. This will be described with reference to FIG. 5 and FIG.

The features learned in different color spaces, that is, the output of hidden layers, can be presented in different feature subspaces, which can provide complementary effects. Hereinafter, the learning DCNN in color space C _k In the specification of the present invention will be described with referred to as DCNN-C _k.

FIG. 3 is an exemplary diagram for explaining a cooperative face color feature learning method according to an embodiment of the present invention, and shows a cooperative face color feature learning framework.

3, the cooperative facial color feature learning framework according to an embodiment of the present invention is derived by the complementary effects of various color spaces in a deep learning framework, such as 1) by DCNN in various color spaces Each color feature extraction, and 2) various color feature fusion using DCNN.

, YIQ, XYZ, HSV, RIQ, RQCr, YQCr 그리고 La*b*을 생성한다. 즉, 입력 얼굴 이미지는 다양한 컬러 공간으로 변형된다. 협동 얼굴 컬러 특징 학습 프레임워크는 생성된 각각의 컬러 공간(DCNN-C_k)에서 DCNN 학습을 통해 각각의 컬러 공간에서의 특징을 담은 정보를 획득할 수 있다. 협동 얼굴 컬러 특징 학습 프레임워크에서 다양한 컬러 특징 융합은 다양한 컬러 공간으로부터 학습된 특징을 한 곳에 모으기 위해 사용된다.The cooperative facial color feature learning framework may include input face images, for example, RGB face images from various color spaces, for example, RGB, YCbCr

, YIQ, XYZ, HSV, RIQ, RQCr, YQCr, and La * b *. That is, the input face image is transformed into various color spaces. The cooperative facial color feature learning framework can obtain information about features in each color space through DCNN learning in each of the generated color spaces (DCNN- _Ck ). Cooperative Face Color Features In the learning framework, various color feature fusions are used to gather learned features from various color spaces.

Here, the convergence feature in which the various color features are fused can be obtained by concatenating and combining all feature vectors from various color spaces in order to obtain a high face recognition rate. However, the discriminative capabilities of the feature vectors in the various color spaces are different, and therefore maximum performance is not guaranteed to simply chain all feature vectors. In addition, a dimensional problem due to an increased dimension may also occur.

The convergence feature in the cooperative face color feature learning framework can be generated or obtained by learning deep neural networks (DNN) composed of a plurality of fully connected layers. Therefore, the cooperative facial color feature learning framework learns and collects feature vectors learned from DCNN-C _k in various color spaces, and the final output feature vector contains only the essential contents, which can improve the discrimination ability of learned features .

, YIQ, XYZ, HSV, RIQ, RQCr, YQCr 그리고 La*b*으로 변형된다. 그리고, 다양한 컬러 공간에서 학습된 DCNN 즉, DCNN-C₁, DCNN-C₂, DCNN-C₃ 등으로 명명되는 DCNN으로부터 d-차원의 다양한 특징 벡터들 y_c1, y_c2, y_c3,... , y_cn이 추출된다. 이러한 모든 특징 벡터들을 연쇄시키면 (nxd)-차원의 특징 벡터 즉, 융합 특징이 형성된다. 즉,

이 된다. 그 후, 앞단에서 나온 y에 대해, Q^(m)과 b^(m)을 이용하여 파라미터가 결정되는 DNN이 결합된 특징 벡터를 아래 <수학식 3>과 같이 얻을 수 있다.4, a color used as an input, for example, an RGB face image may be represented by various color spaces such as RGB, YCbCr

, YIQ, XYZ, HSV, RIQ, RQCr, YQCr and La * b *. In addition, DCNNs learned in various color spaces, DCNN-C ₁ , DCNN-C ₂ , and DCNN-C ₃ Dimensional feature vectors y _c1 , y _c2 , y _c3 , ..., y _cn are extracted from the DCNN, By concatenating all of these feature vectors, (nxd) -dimensional feature vectors, that is, fusion features are formed. In other words,

. Thereafter, a feature vector to which DNNs whose parameters are determined by using Q ^(m) and b ^(m) with respect to y from the front end is combined can be obtained as Equation (3) below.

&Quot; (3) &quot;

Here, Q ^(m) and b ^(m) denote the weights and weights of the mth layer of DNN, M denotes the number of layers in DNN, f ^(m) , And f ⁽⁰⁾ = y.

The feature vectors concatenated by the DNN parameter can be mapped to a mixed feature space of low dimension and mixed color features from various color spaces can describe the input information in compressed form. Furthermore, in the case of discernible feature vectors, it can be emphasized by the learned weight of DNN.

As described above, the cooperative facial color feature learning framework according to the embodiment of the present invention can enhance the face recognition rate by using the fused features by fusing the extracted color features in various color spaces through learning.

That is, the present invention can recognize the face of the face image using the convergence feature of the face image learned by the cooperative face color feature learning framework.

In other words, another embodiment of the present invention is a method of extracting color features in various color spaces for a face image, generating color features based on learning of color features in the extracted color spaces, , The face recognition rate can be improved by recognizing the face of the face image based on the generated fusion characteristic.

The present invention can prove the face recognition rate through the following experiment.

, YIQ, XYZ, HSV, RIQ, RQCr, YQCr, La*b* 그리고 흑백을 생성한다. 이 때, 실험을 위하여, 조도 변화 아래서 촬영된 Multi-PIE dataset을 사용하며, Dataset은 200개 주제에 대해 14320개의 트레이닝 이미지(training image), 5863개의 프로브(probe) 얼굴 이미지, 137개 주제로부터 137 갤러리 얼굴 이미지(총 주제의 수는 337개)로 구성될 수 있다.First, in order to prove the present invention, nine different color spaces YCbCr

, YIQ, XYZ, HSV, RIQ, RQCr, YQCr, La * b *, and black and white. At this time, for the experiment, we use Multi-PIE dataset photographed under illumination change. Dataset has 14320 training images, 5863 probe face images for 200 subjects, 137 And a gallery face image (the total number of subjects is 337).

All facial images are resized to 32x32 pixels. The subject information (subject or ID) used in training and testing is different from each other. The face recognition rate is calculated from the feature vector extracted from the face image of the gallery, 1 - Nearest neighbor classification.

In the DCNN architecture of the present invention, the learning rate is set to 0.01 and attenuated to an exponential function, with a probability of dropout of 0.5 at the Softmax layer and 0.25 at the two subsampling layers. In the DNN structure, the two fully connected layers consist of 2048 units, 512 units and a Softmax layer. In DNN, the learning rate is 0.001 and the dropout probability is set to 0.5.

일 수 있다. 또한, DCNN과 DNN을 모두 학습하기 위해서 크로스 엔트로피 로스(cross-entropy loss)함수가 사용된다.And in DCNN-C _k, activation function σ (x) for the final Softmax layer may be a sigmoid (sigmoid) and function, σ (x) = 1 / (1-e -x). ReLU can be used as an activation function in other layers. In other words,

Lt; / RTI &gt; In addition, a cross-entropy loss function is used to learn both DCNN and DNN.

Table 1 below shows the FR accuracy of DCNN-C _k learned in various color spaces.

얼굴 이미지의 경우 흑백 이미지와 비교하여 얼굴인식률은 17.05% 더 높고 RGB 얼굴 인식과 비교했을 때 8.27% 높은 것을 알 수 있다. 이는 얼굴 신분 정보가 컬러 공간에 따라 DCNN-C_k에서 다르게 학습됨을 보이고 있다.As shown in Table 1, the face recognition rate of the color face image is higher than that of the black and white face image, and La * b *

The face recognition rate is 17.05% higher than the monochrome image and 8.27% higher than the RGB face recognition. This shows that face identification information is learned differently in DCNN-C _k according to color space.

에 의해 서로 다른 컬러 공간에 서로 다른 모습으로 나타난다. 뿐만 아니라 도 7에 도시된 바와 같이, 특징 서브 공간에 있는 Class1(RGB)와 Class1(Y

CbCr) Class1에 대해 RGB와 YCbCr 특징을 살펴보면 같은 ID를 가짐에도 불구하고 매우 멀리 떨어져 있으며, 심지어 특징 서브 공간에서 클래스간 거리를 넘기도 하는 것을 알 수 이다.For example, FIGS. 5 and 6 show the test face image when the first two convolution layers are passed. The face identification characteristic map is DCNN-C _k

In different color spaces. In addition, as shown in FIG. 7, in the feature subspace, Class1 (RGB) and Class1 (Y

CbCr) Looking at the RGB and YCbCr features for Class1, we can see that even though they have the same ID, they are very distant, even beyond the distance between classes in feature subspaces.

, YIQ, XYZ, HSV, RIQ, RQCr, YQCr 그리고 La*b*

컬러 공간으로부터 338, 326, 342, 326, 341, 333, 327, 357, 310개의 특징들로 구성되며, 각각의 컬러 공간에서 기여된 특징들은 약 10퍼센트 정도로 균등한 비율로 존재한다. To analyze the complementary effects of each feature in various color spaces, only some features are extracted from 4,608 (512-dimensional x 9-color space) dimensional feature vectors based on Fisher's criterion, and 4,608 dimensional feature vectors When 3,000 feature vectors are selected, the face recognition rate is about 87%. The 3,000 feature vectors are RGB, YCbCr

, YIQ, XYZ, HSV, RIQ, RQCr, YQCr, and La * b *

The color space consists of 338, 326, 342, 326, 341, 333, 327, 357, and 310 features from the color space, and the features contributed in each color space are present in equal proportions on the order of about 10 percent.

가 사용된 얼굴 인식 성능 결과로부터 다양한 컬러 공간에서 학습된 특징이 얼굴 인식 결과에서 상호 보완적인 효과를 발생시키는 것을 알 수 있다.As shown in Table 1, one DCNN- _Ck

The results show that the features learned in various color spaces generate complementary effects in the face recognition results.

Table 2 below shows the comparison of face recognition performance using the method presented in the Multi-PIE dataset (collaborative feature learning) and face recognition performance using the feature-level concatenation method.

As described above, the face recognition accuracy increases to 87.34% when the features obtained in the nine color spaces are used in a cascade connection. By merely fusing the various features, highly discernible results can be obtained from different color spaces. When the feature fusion method given in DNN is used, the face recognition accuracy of a characteristic fused using complementary information from multiple DCNN-C _k, removing features further not important, and can be improved at a low characteristic dimension Learning can be carried out.

As described above, it can be seen that the method according to the embodiments of the present invention has a much higher face recognition rate than the monochrome face recognition in the color face image, and the face recognition rate is improved according to the selection and fusion of the learned features in various color spaces Able to know. In particular, the DNN based on the fused features makes it possible to fully utilize complementary information existing in various color spaces, and as a result, the face recognition rate can be further improved.

FIG. 7 illustrates a configuration of a cooperative facial color feature learning apparatus according to an embodiment of the present invention, and shows a configuration of an apparatus for performing the cooperative facial color feature learning method.

Referring to FIG. 7, the cooperative facial color feature learning apparatus 700 according to the embodiment of the present invention includes an extraction unit 710 and a fusion unit 720.

The extracting unit 710 extracts color features from each of the predetermined plurality of color spaces with respect to the face image.

At this time, the extracting unit 710 can extract the color characteristic of each of the color spaces by learning the characteristic of the DCNN of each of the color spaces.

The contents of such an extractor may also include all contents extracting color features in color spaces in the above-described method.

The fusion unit 720 generates a fusion feature in which the extracted color features are fused based on learning of the extracted color features.

At this time, the fusion unit 720 can generate fusion characteristics by concatenating all the feature vectors for the extracted color features.

At this time, the fusion unit 720 can generate convergence characteristics by learning the DNN for the extracted color features, and the DNN can be composed of a plurality of completely connected layers.

Of course, the contents of the fusion unit may include all the contents that generate or output the fusion characteristic by fusing the color features extracted in the above-described method.

FIG. 8 shows a configuration of a face recognition apparatus using a facial color feature according to an embodiment of the present invention. FIG. 8 illustrates a face recognition apparatus using a fused feature fused by the facial color feature learning apparatus of FIG. FIG.

Referring to FIG. 8, the face recognition apparatus 800 according to the embodiment of the present invention includes an extraction unit 810, a fusion unit 820, and a recognition unit 830.

The extracting unit 810 extracts color features from each of the predetermined plurality of color spaces with respect to the face image.

At this time, the extracting unit 810 can extract the color characteristic of each of the color spaces by learning the characteristic of the DCNN of each of the color spaces.

The contents of such an extractor may also include all contents extracting color features in color spaces in the above-described method.

The fusion unit 820 generates a fusion feature in which the extracted color features are fused based on learning of the extracted color features.

At this time, the fusion unit 820 can generate fusion characteristics by concatenating all the feature vectors for the extracted color features.

At this time, the fusion unit 820 can generate convergence characteristics by learning the DNN for the extracted color features, and the DNN can be composed of a plurality of completely connected layers.

Of course, the contents of the fusion unit may include all the contents that generate or output the fusion characteristic by fusing the color features extracted in the above-described method.

The recognition unit 830 recognizes the face of the face image based on the generated fusion characteristic.

The system or apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the systems, devices, and components described in the embodiments may be implemented, for example, as processors, controllers, arithmetic logic units (ALUs), digital signal processors, microcomputers, field programmable gate arrays such as an array, a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to embodiments may be implemented in the form of a program instruction that may be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI &gt; or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (16)

Extracting color features from each of a plurality of predetermined color spaces, for a face image; And
Generating a fused feature wherein the extracted color features are fused based on learning of the extracted color features
A face color feature learning method.
The method according to claim 1,
The step of extracting the color feature
Wherein the color features are extracted by learning features of deep convolutional neural networks (DCNN) of each of the color spaces.
The method according to claim 1,
The step of generating the fusion feature
Wherein the convergence feature is generated by concatenating all the feature vectors for the extracted color features.
The method according to claim 1,
The step of generating the fusion feature
And learning the DNN (deep neural networks) for the extracted color features to generate the fusion feature.
5. The method of claim 4,
The DNN
And a plurality of fully connected layers.
Extracting color features from each of a plurality of predetermined color spaces, for a face image;
Generating a fusion feature in which the extracted color features are fused based on learning of the extracted color features; And
Recognizing a face of the face image based on the generated fusion characteristic
Wherein the facial color feature is a face image.
The method according to claim 6,
The step of extracting the color feature
Wherein the color features are extracted by learning characteristics of deep convolutional neural networks (DCNN) of each of the color spaces.
The method according to claim 6,
The step of generating the fusion feature
Wherein the convergence feature is generated by learning DNNs (deep neural networks) composed of a plurality of fully connected layers with respect to the extracted color features.
An extraction unit for extracting color features from each of the predetermined plurality of color spaces, with respect to the face image; And
Wherein the extracted color features are based on learning of the extracted color features,
Wherein the face color feature learning device comprises:
10. The method of claim 9,
The extracting unit
And extracts the color features by learning features of deep convolutional neural networks (DCNN) of each of the color spaces.
10. The method of claim 9,
The fusion unit
And the fusion feature is generated by concatenating all the feature vectors for the extracted color features.
10. The method of claim 9,
The fusion unit
And the convergence feature is generated by learning DNN (deep neural networks) for the extracted color features.
13. The method of claim 12,
The DNN
And a plurality of fully connected layers.
An extraction unit for extracting color features from each of the predetermined plurality of color spaces, with respect to the face image;
A fusion unit for generating a fusion feature in which the extracted color features are fused based on learning of the extracted color features; And
A recognition unit for recognizing a face of the face image based on the generated fusion characteristic,
Wherein the facial color feature includes at least one of a face image and a face image.
15. The method of claim 14,
The extracting unit
Wherein the color features are extracted by learning characteristics of deep convolutional neural networks (DCNN) of each of the color spaces.
15. The method of claim 14,
The fusion unit
And the convergence feature is generated by learning DNNs (deep neural networks) composed of a plurality of fully connected layers with respect to the extracted color features.

Images