JP2001338290A - Device and method for image processing and computer- readable with medium recording recorded with image processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve recognition rate of an image. SOLUTION: This device is provided with an image input part 102 for inputting a color image, a color model generating part 201 for generating a color model by applying a statistic method to the inputted color image, and subjecting a model image to color conversion of the inputted color image on the generated color model, and a feature model generating part 202 for generating a feature space by applying a statistical method to plural color model images converted by the color model generating part 201.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an image processing apparatus,
The present invention relates to a computer-readable recording medium storing an image processing method and an image processing program, and more particularly, to an image processing apparatus, an image processing method, and an image processing program for recognizing an object in consideration of color information unique to the object to be recognized. The present invention relates to a recorded computer-readable recording medium.

[0002]

2. Description of the Related Art Techniques for recognizing a subject from an image obtained by imaging the subject have been developed. JP 5
Japanese Patent No. -20442 discloses an eigenspace method for performing pattern recognition on a low-dimensional eigenspace obtained by performing a transformation such as principal component analysis on a pattern of a high-dimensional grayscale image including a human face. A face image collating device for recognizing a human face is described. In this technique, a recognition model is generated using a grayscale grayscale image obtained by capturing a face image of a person with a camera. The generation of the recognition model is obtained by converting a pattern of a high-dimensional gray-scale image using principal component analysis. The required recognition model is represented by a low-dimensional eigenspace, and is a method of recognizing a face image of a target person by performing pattern recognition on the low-dimensional eigenspace.

[0003]

However, the technique described in Japanese Patent Application Laid-Open No. H5-204442 generates a recognition model only from the grayscale information and the shape of an image using a grayscale grayscale image. The color characteristics of the person to be recognized are not included in the recognition model. For this reason,
There were limits to improving recognition accuracy.

A color image can be used to generate a recognition model.
In the case of three images of green (G) and blue (B), the amount of image data is tripled as compared with the case of using a gray image. For this reason, there has been a problem that time is required for processing.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems. One of the objects of the present invention is to easily generate a space that sufficiently represents the color distribution of an input color image. An object of the present invention is to provide an image processing apparatus, an image processing method, and a computer-readable recording medium in which an image processing program is recorded, in which an image recognition rate is improved.

[0006]

According to one aspect of the present invention, there is provided an image processing apparatus comprising: an input unit for inputting a color image; and a statistical method for inputting the color image. A color model space generating means for generating a color model space by applying a color model space, and a color converting means for performing color conversion of the input color image into a model image on the generated color model space.
A feature space generation unit configured to generate a feature space by applying a statistical method to the plurality of model images converted by the color conversion unit.

According to the present invention, a color model space is generated by applying a statistical method to an input color image, and the input color image is color-converted into a model image in the generated color model space. You. Also, a statistical method is applied to the plurality of color-converted model images to generate a feature space. The model image in the color model space is an image sufficiently representing the color characteristics of the input image. The feature space is a space that sufficiently represents the color distribution of the input color image. As a result, a space that sufficiently represents the color distribution of the input color image can be easily generated, and an image processing apparatus with an improved image recognition rate can be provided.

Preferably, the color model space generating means of the image processing apparatus performs a principal component analysis on the input color image to convert a color space lower in dimension than the color space of the input color image into a color model. It is characterized by being generated as a space.

According to the present invention, since a principal component analysis is performed on a color image, a low-dimensional color model space can be easily generated.

[0010] More preferably, the feature space generating means of the image processing apparatus generates a feature space by performing principal component analysis on a plurality of color-converted model images.

According to the present invention, principal component analysis is performed on a plurality of color-converted model images, so that a low-dimensional feature space can be easily generated.

According to another aspect of the present invention, an image processing apparatus includes an input unit for inputting a color image, and a color model space that is a lower-dimensional color space than the color space of the color image. Color conversion means for performing color conversion to the above model image, and feature space generating means for generating a feature space by applying a statistical method to the plurality of model images color-converted by the color conversion means.

According to the present invention, an input color image is color-converted into a model image in a color model space, and a statistical method is applied to the plurality of color-converted model images to generate a feature space. . For this reason, the data amount can be reduced by reducing the dimension of the color image. When the color model space is a general-purpose space irrespective of a color image, the model image is an image sufficiently representing the color characteristics of the input image. As a result, a space that sufficiently represents the color distribution of the input color image can be easily generated, and an image processing apparatus with an improved image recognition rate can be provided.

[0014] Preferably, the color model space of the image processing apparatus is generated by performing principal component analysis on a plurality of color images.

According to the present invention, since a principal component analysis is performed on a color image, a low-dimensional color model space can be easily generated.

[0016] More preferably, the image processing apparatus includes: a projecting unit for projecting the model image converted by the color converting unit onto the generated feature space; and a color conversion unit for the predetermined color image input by the input unit. After the color conversion by the means, the storage means for storing the dictionary vector obtained by projecting into the feature space by the projecting means, and the color image newly input by the input means, after the color conversion by the color conversion means, the projection means The image processing apparatus further includes feature vector calculating means for projecting the feature vector to obtain a feature vector, and comparing means for comparing the obtained feature vector with a stored dictionary vector.

According to the present invention, after a predetermined color image is color-converted, a dictionary vector obtained by projecting the color image into a feature space is stored, and after a newly input color image is color-converted, the dictionary vector is converted into a feature space. The projection is performed to obtain a feature vector, and the obtained feature vector is compared with the stored dictionary vector. Since the predetermined color image and the newly input image are compared in the feature space, it is possible to provide an image processing apparatus with an improved image recognition rate.

More preferably, the image processing apparatus is characterized in that the predetermined color image is a color image used for obtaining a feature space by the feature space generation unit.

According to the present invention, since the predetermined color image is a color image used for obtaining the feature space in the feature space generation unit, the image can be recognized more accurately.

[0020] More preferably, the image processing apparatus further comprises extraction means for extracting a face area from the input color image.

According to the present invention, a face area is extracted from an input color image. Therefore, it is possible to provide an image processing apparatus in which the recognition rate of a human face is improved.

According to another aspect of the present invention, an image processing method includes a step of inputting a color image, a step of performing a statistical method on the input color image to generate a color model space, and The method includes the steps of color-converting a color image into a model image in a generated color model space, and generating a feature space by applying a statistical method to the plurality of converted model images.

According to the present invention, the input color image is subjected to a statistical method to generate a color model space, and the input color image is color-converted into a model image in the generated color model space. You. Further, a statistical method is applied to the plurality of converted model images to generate a feature space.
The model image in the color model space is an image sufficiently representing the color characteristics of the input image. The feature space is
This is a space that sufficiently represents the color distribution of the input color image. As a result, a space that sufficiently represents the color distribution of the input color image can be easily generated, and an image processing method with an improved image recognition rate can be provided.

According to another aspect of the present invention, a computer-readable recording medium includes: a step of inputting a color image; a step of performing a statistical method on the input color image to generate a color model space; Causing the computer to execute a step of performing color conversion of the input color image into a model image on the generated color model space and a step of performing a statistical method on the plurality of converted model images to generate a feature space; The image processing program of the above is recorded.

According to the present invention, a color model space expressing color characteristics is generated by applying a statistical method to the input color image, and the color model space on the color model space where the input color image is generated is generated. The color is converted to a model image. Further, a statistical method is applied to the plurality of converted model images to generate a feature space. The model image in the color model space is
This is an image that sufficiently represents the color characteristics of the input image. The feature space is a space that sufficiently represents the color distribution of the input color image. As a result, it is possible to provide a computer-readable recording medium in which an image processing program which can easily generate a space sufficiently representing a color distribution of an input color image and has an improved image recognition rate is recorded. Can be.

[0026]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an object recognition apparatus according to the present embodiment will be described with reference to the drawings. The same reference numerals in the drawings denote the same or corresponding members, and description thereof will not be repeated.

FIG. 1 is a block diagram showing a schematic configuration of an object recognition apparatus according to one embodiment of the present invention. Referring to FIG. 1, an object recognition device 100 includes a camera 101
And the external storage device 111.

The object recognition apparatus 100 includes an image input unit 102 connected to the camera 101, a learning unit 300 connected to the image input unit 102, a recognition processing unit 400 connected to the image input unit 102, A control unit 110 for controlling the entire device 100;

The camera 101 is a CCD (Charge Coupled).
This is a camera that includes a solid-state image sensor such as a device and outputs a full-color image signal by imaging a subject. The camera 101 captures an image of a subject and outputs RGB image signals to the image input unit 102.

The image input unit 102 converts an image signal input from the camera 101 into digital image data represented by, for example, RGB, and corrects luminance values, cuts out a recognition target area, and normalizes its shape. After processing
The image data subjected to the correction processing is output to the learning unit 300 or the recognition processing unit 200.

The learning unit 300 includes a color model generation unit 201 for generating a color model based on color data included in a plurality of image data, a color model storage unit 301 for storing the generated color model, A feature model generation unit 202 for generating a feature model based on a plurality of image data projected on the model space of the generated color model; a feature model storage unit 302 for storing the feature model; And a dictionary vector storage unit 303 for storing the obtained color model image data as a dictionary vector. Note that the color model image data is data obtained by projecting the image data into the model space of the color model.

The color model generation unit 201 includes the image input unit 10
The eigenspace is generated by performing a principal component analysis on the pixel values of the image data input from Step 2. Color model generator 20
The eigenspace generated in step 1 is called a color model. The color model is represented by an eigenvector obtained by principal component analysis, and the eigenvector is stored in the color model storage unit 301. At this time, when the input image data is represented by three of RGB, if the color model is a space represented by one dimension, the input image data is compressed into one-dimensional data.

The feature model generation unit 202 performs statistical processing such as principal component analysis using a plurality of color model image data obtained by projecting image data on the color model generated by the color model generation unit 201. Generates an eigenspace. The eigenspace generated by the feature model generation unit 202 is
It is called a feature model. The feature model is defined by eigenvectors obtained by principal component analysis. The eigenvector to be obtained is obtained until the contribution ratio reaches a predetermined value. Thereby, the dimension of the image data is compressed, and the data amount can be reduced. The eigenvector defining the feature model is stored in the feature model storage unit 302.

On the other hand, the color model image data used for generating the feature model in the feature model generation unit 202 is projected on the feature model. The parameters output by the projection are stored in the dictionary vector storage unit 303 as dictionary vectors.

As the color model image data used for generating the feature model in the feature model generation unit 202, a plurality of image data obtained by imaging the same person by the camera 101 may be used, and the image data may be obtained by imaging the faces of a plurality of people. Images may be used. When generating a feature model using a plurality of image data obtained by imaging the face of the same person, a dictionary vector obtained by projecting the color model image data onto the feature model is:
They are classified into the same category and stored in the dictionary vector storage unit 303.

As described above, the learning unit 300 generates a color model by performing principal component analysis on color information of full-color image data obtained by imaging a person, and stores the image in the model space of the generated color model. By projecting the data, the dimension of the image data is compressed. Further, the generated color model takes into account the color information of the input image data, so that it becomes an eigenspace that sufficiently takes into account the distribution of colors contained in the image. In particular, in the present embodiment,
Since the input image data is image data including a person's face, an eigenspace that sufficiently takes into account the color distribution of the person's face is generated. The color model image data on which the image data is projected onto the eigenspace is data in which the color distribution is sufficiently considered.

The feature model generation unit 202 performs a principal component analysis on color model image data obtained by projecting the input full-color image data onto a color model, and generates a feature model. Dimensions can be reduced and processing speed is increased. In particular, when the color model generated by the color model generation unit 201 is defined only by the first principal component, the input image data is RGB.
From three dimensions to one-dimensional data of the first principal component.

A recognition processing unit 400 extracts a color conversion unit 401 for performing color conversion on image data input from the image input unit 102 and a feature vector of the color model image data color-converted by the color conversion unit 401. And a matching unit 403 for comparing the extracted feature vector with a dictionary vector stored in the dictionary vector storage unit 303.

The color conversion unit 401 performs color conversion by projecting the image data input from the image input unit 102 onto the color model stored in the color model storage unit 301. Color conversion is performed using eigenvectors that define a color model. Therefore, when the eigenvectors that define the color model are only the eigenvectors based on the first principal component, the input image data represented in three-dimensional RGB is a one-dimensional color model image of only the first principal component. Converted to data. This will be described later.

The feature vector extraction unit 402
The feature vector is extracted by projecting the color model image data color-converted in step 01 to the feature model stored in the feature model storage unit 302. Extraction of a feature vector is performed using an eigenvector that defines a feature model.

The collating unit 403 includes a feature vector extracting unit 40
2 is stored in the dictionary vector storage unit 3
The dictionary vector having the highest correlation is obtained by comparing with the dictionary vector stored in No. 03. Dictionary vector storage unit 30
In the case where the dictionary vectors stored in No. 3 are stored for each category, the input image data is classified into the category having the highest correlation with the feature vector. Thereby, it is possible to recognize who the person included in the input image data is.

The external storage device 111 includes a magneto-optical disk,
Recording medium 112 such as a digital video disk (DVD)
This is a driver for reading programs and data stored in the.

The learning unit 300 and the recognition processing unit 40
0 is recorded on the recording medium 112, and the recorded object recognition program is read by the external storage device 111, whereby the control unit 1
10 may be executed. In this case, the learning unit 300 and the recognition processing unit 400 are unnecessary, and the control unit 110 stores a color model, a feature model, and a dictionary vector in a storage unit (not shown).

Next, the flow of processing performed by the object recognition apparatus 100 will be specifically described. FIG. 2 is a flowchart showing a flow of a learning process performed by learning section 300 of object recognition device 100 in the present embodiment. Referring to FIG. 2, in the learning process, first, image data is input (step S1). Image data is input to the image input unit 1
02 is the data represented by the RGB signal input from the device.

In the present embodiment, the RGB signals are used as the color image signals. However, the present invention is not limited to this. The image signals Y, I and I, which are composed of the luminance signal Y and the two color difference signals Cb and Cr, are used. , Q signals, Y, U, V signals, or H, V, C signals in a uniform visual perception space such as a Munsell space.

Next, a color model is generated by performing a principal component analysis on the pixel values of the input image data. Here, a case where the number of input image data is n will be described. here,
Assuming that one image data includes three images of an R image, a G image, and a P image, and that each image includes N pixels, n image data is represented by Expression (1). It can be represented by a matrix D. In matrix D, R _i1
R _i2 ... R _iN indicate pixel values included in the i-th image. For this reason, the matrix D is represented by a matrix of 3 rows × (N × n) columns.

To generate a color model, a principal component analysis is performed on the matrix D. Therefore, the variance-covariance matrix C of the matrix D
Is calculated based on the equation (2). Then, the obtained variance-covariance matrix C is diagonalized by an appropriate method, and an eigenvector matrix E of the variance-covariance matrix C and an eigenvalue vector e corresponding to each eigenvector are calculated.

The obtained eigenvector matrix E is given by (3)
The eigenvalue vector F is expressed by Expression (4).

At this time, if the color deviation of the input image is large, the eigenvalue becomes ω ₁ >> ω ₂ >> ω ₃ .
Eigenvector W _{1 of} the first principal component corresponding to eigenvalue ω ₁ =
A color model based on only (e ₁₁ , e ₁₂ , e ₁₃ ) can be defined. Note that the eigenvalues only with the eigenvectors W ₁ corresponding to omega _1, when the contribution rate is not sufficient, the eigenvalues eigenvectors of the second principal component corresponding to omega ₂ W ₂ = (e
_21, e _22, e ₂₃₎ may be defined a color model with base also. In this case, RGB three-dimensional image data is represented by two-dimensional image data represented by a first principal component and a second principal component.

The base vector of the color model is stored in the color model storage unit 301 as a color conversion parameter. Here, the eigenvector W ₁ of the first principal component is set as an eigenvector serving as a basis of the color model, and the eigenvector components e ₁₁ ,
e ₁₂ and e ₁₃ are color model storage units 3 as color conversion parameters.
01 is stored.

Next, all of the n pieces of image data used for generating the color model by the color model generation unit 201 are projected onto the color model to perform color conversion (step S4). Color conversion is
This is performed using the color conversion parameters stored in the color model storage unit 301. More specifically, the image input unit 102
Assuming that the RGB signals of one pixel of the input image data are R, G, and B, the color-converted value T is (5)
It is represented by the formula. As described above, the image data of the RGB signal is color-converted into the color model image data of the pixel value T after the conversion.

[0052]

(Equation 1)

Then, principal component analysis is performed on the converted color model image data to generate a feature model (step S5). In the generation of the feature model, principal component analysis is performed until the cumulative contribution rate when the eigenspace is formed reaches a predetermined value. That is, a plurality of eigenvectors are obtained by performing principal component analysis. Among the plurality of eigenvectors, eigenvectors of a first principal component, a second principal component, a third principal component... Is done. In this manner, the eigenvectors are obtained until the cumulative contribution rate reaches a predetermined value. Then, a feature model is defined by the obtained eigenvectors.

The obtained eigenvectors are stored in the feature model storage unit 302 as image conversion parameters (step S6).

The color model image data corresponding to the recognition target image among the color model image data subjected to color conversion in step S4 is projected on the feature model (step S7). The projection is performed using the image conversion parameters stored in the feature model storage unit 302. Then, the vector obtained by the projection in step S7 is stored in the dictionary vector storage unit 303 as a dictionary vector (step S8). Here, the recognition target image refers to image data including a person to be subjected to image recognition. When the recognition target image includes a plurality of images of the same person as a subject, a plurality of dictionary vectors are obtained by projecting a plurality of color model image data corresponding to the plurality of recognition target images onto a feature model. The plurality of obtained dictionary vectors are classified into the same category. When the persons included in the two image data are not the same person, the dictionary vectors obtained from the two image data are classified into different categories and stored in the dictionary vector storage unit 303.

In the present embodiment, the image used to create the feature model is the same as the image used to create the color model. However, if a general-purpose color model applicable to any image set has already been created, there is no need to create a color model for the image set. it can. In this case, an image used to create a general-purpose color model can be different from an image used to create a feature model.

FIG. 3 is a flowchart showing the flow of a recognition process performed by the recognition processing section 400 of the object recognition device 100 according to the present embodiment. Referring to FIG. 3, in the recognition process, first, image data to be processed is input from image input unit 102 (step S11). at this point,
It is unknown which person is included in the image data to be processed.

The color conversion is performed by projecting the input image data onto the color model stored in the color model storage section 301 (step S12). The projection onto the color model is performed using equation (5).

Then, the color-converted color model image data is projected onto the feature model stored in the feature model storage unit 302 (step S13). The projection to the feature model is performed using the eigenvectors that define the feature model stored in the feature model storage unit 302. As a result, a feature vector corresponding to the input image is obtained.

The obtained feature vector is compared with the dictionary vector stored in the dictionary vector storage unit 303. For the comparison, a difference between vectors may be obtained, or a Euclidean distance or a Mahalanobis distance may be used. When the dictionary vectors are classified into categories, a plurality of dictionary vectors included in the category are compared with the feature vectors.

Based on the comparison result in step S14, it is determined in step S15 whether or not the feature vector and the dictionary vector are correlated. If there is a correlation, the process proceeds to step S16. Step S14
Are compared with other dictionary vectors. The presence / absence of the correlation is determined, for example, as having a correlation when the Euclidean distance is within a predetermined range, and as having no correlation otherwise.
If no correlated dictionary vector is found even after comparison with all dictionary vectors, the processing is terminated as unrecognizable.

In step S16, the input image is classified into a category including a correlated dictionary vector. Thereby, it is possible to recognize who the person included in the input image is.

As described above, in the object recognition apparatus according to the present embodiment, the color characteristics of an object are analyzed by using principal component analysis, and an image is represented by color parameters incorporating the color characteristics to the maximum. I have to. For this reason, it is possible to generate a feature model with high recognition accuracy and capable of recognizing an object in low dimensions.

Since a low-dimensional feature model is used when actually performing the recognition processing, the object recognition apparatus can be simplified and the processing can be performed at a high speed.

The case where the object recognition apparatus of this embodiment generates a feature model using a color image of a human face, and the case where a feature model is generated by performing principal component analysis using a grayscale image of a human face. The cumulative contribution ratio was compared with the number of input color images or grayscale images being 300. The result is shown in FIG. FIG. 4 is a diagram for describing an effect of the object recognition device according to the present embodiment.

Referring to FIG. 4, according to the conventional method using the grayscale image, in order to obtain a cumulative contribution rate of 50%, six dimensions (sixth principal component) from the one with the largest variance are required. On the other hand, according to the object recognition device of the present invention, the cumulative contribution ratio exceeds 50% when three-dimensional (third principal component) data, which is about half that of the conventional method, is used. Therefore, according to the object recognition device of the present embodiment, a sufficient contribution ratio can be obtained in a lower dimension than in the conventional method.

In this embodiment, the object to be recognized is a human face. However, the present invention is not limited to this, and it goes without saying that other objects can be recognized. If a face area is extracted from an image by an appropriate method, and the above-described learning processing and recognition processing are performed only on the extracted image area, the object can be recognized with higher accuracy. In this case, human skin,
Since colors can be expressed using parameters specialized for hair, lips, and the like, gender, age, and facial expression can be accurately determined, and personal authentication and the like can be performed effectively.

Further, in the object recognition apparatus according to the present embodiment, the principal component analysis is used for generating the color model and the feature model. However, the present invention is not limited to this, and an appropriate method such as orthogonal transformation can be used.

Although the object recognition apparatus has been described in the present embodiment, an object recognition method for executing the processes shown in FIGS. 2 and 3 or an object recognition program for causing a computer to execute these processes is provided. Needless to say, the invention can be regarded as a computer-readable recording medium on which the information is recorded.

The embodiments disclosed this time are to be considered in all respects as illustrative and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a schematic configuration of an object recognition device according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating a flow of a learning process performed by a learning unit of the object recognition device according to the present embodiment.

FIG. 3 is a flowchart illustrating a flow of a recognition process performed by a recognition processing unit of the object recognition device according to the present embodiment.

FIG. 4 is a diagram for describing an effect of the object recognition device according to the present embodiment.

[Explanation of symbols]

Reference Signs List 100 object recognition device, 101 camera, 102 image input unit, 110 control unit, 111 external storage device, 11
2 recording medium, 200 recognition processing unit, 201 color model generation unit, 202 feature model generation unit, 300 learning unit, 3
01 color model storage, 302 feature model storage, 3
03 dictionary vector storage unit, 400 recognition processing unit, 401
Color conversion unit, 402 Feature vector extraction unit, 403 Collation unit.

Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat II (Reference) H04N 1/46 H04N 1/46 Z F Term (Reference) 5B043 AA09 BA04 DA09 EA18 GA04 5B057 AA20 BA02 CA01 CA08 CA12 CA16 CE17 DA12 DC19 DC25 DC34 5C079 HB01 LB02 MA11 NA27 NA29 5L096 AA02 AA06 BA20 CA02 DA02 FA15 FA32 FA34 FA38 FA46 GA38 GA40 JA11 JA22 KA15

Claims (10)

[Claims] An input unit for inputting a color image; a color model space generating unit configured to generate a color model space by applying a statistical method to the input color image; Color conversion means for performing color conversion to a model image on a color model space, and a feature space generation means for generating a feature space by performing a statistical method on the plurality of model images converted by the color conversion means, Image processing device. 2. The color model space generating means performs principal component analysis on an input color image,
2. The color model space according to claim 1, wherein a color space having a lower dimension than the color space of the input color image is generated.
An image processing apparatus according to claim 1. 3. The image processing apparatus according to claim 1, wherein said feature space generation means generates a feature space by performing principal component analysis on a plurality of color-converted model images. . 4. An input unit for inputting a color image, and a color conversion unit for converting the input color image into a model image on a color model space that is a lower-dimensional color space than the color space of the color image. An image processing apparatus comprising: a feature space generation unit configured to generate a feature space by applying a statistical method to a plurality of model images color-converted by the color conversion unit. 5. The image processing apparatus according to claim 4, wherein the color model space is generated by performing principal component analysis on a plurality of color images. 6. A projecting means for projecting the model image transformed by the color transforming means onto the generated feature space, and a color of the predetermined color image inputted by the inputting means is inputted by the color transforming means. After the conversion, a storage unit for storing a dictionary vector obtained by projecting the feature space by the projecting unit, and a color image newly input by the input unit is subjected to color conversion by the color conversion unit, and then the projection is performed. 2. A feature vector calculating means for projecting the feature vector to the feature space to obtain a feature vector, and a comparing means for comparing the obtained feature vector with the stored dictionary vector.
The image processing apparatus according to any one of claims 1 to 5, 7. The image processing apparatus according to claim 6, wherein the predetermined color image is a color image used for obtaining a feature space in the feature space generation unit. 8. The image processing apparatus according to claim 1, further comprising an extraction unit configured to extract a face region from the input color image. 9. A step of inputting a color image, a step of performing a statistical method on the input color image to generate a color model space, and a step of converting the input color image into the generated color model space. An image processing method, comprising the steps of: performing a color conversion into a model image, and performing a statistical method on the converted plurality of model images to generate a feature space. 10. A step of inputting a color image, a step of performing a statistical method on the input color image to generate a color model space, and a step of converting the input color image into the generated color model space. A computer-readable recording in which an image processing program for causing a computer to execute a step of performing color conversion into a model image and a step of performing a statistical method on the plurality of converted model images to generate a feature space is recorded. Medium.