JP2014093043A - Face image analyzer and face image analysis method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique for highly accurate quantization of the quality of the entire face based on a face image.SOLUTION: A face analyzer obtains average face image information and a plurality of basis functions relating to a plurality of sample face images normalized in terms of shape, and analyzes the principal component of an analysis target face image normalized in face, thereby calculating each of the basis coefficients for the analysis target face image. On the basis of each basis coefficient, the basis functions and average face image information, the face analyzer forms each of reproduced face images corresponding to the next cumulative contribution rate for a prescribed accuracy, and calculates the differences between the reproduced face images in terms of brightness and darkness information, thereby obtaining at least the brightness difference image or darkness difference image for each of the pair of adjacent cumulative contribution rates. The face image analyzer calculates at least a brightness component central value or a darkness component central value for each of the pairs of adjacent cumulative contribution rates. The face image analyzer calculates at least a brightness component cumulative value or darkness component cumulative value for each of the pairs of adjacent cumulative contribution rates.

Description

  The present invention relates to a technique for analyzing a face image.

  Various methods have been proposed in which a specific part of the skin is measured using an image analysis technique or the like, and the skin texture such as transparency and gloss is objectively quantified based on the measurement result.

  Patent Document 1 below proposes a method for evaluating the translucency of skin by making P-polarized light and S-polarized light incident on the skin surface and calculating the reflectance of the reflected light. Patent Documents 2 and 3 below propose a method for evaluating the state of skin and the degree of makeup collapse based on the shape of a histogram obtained by measuring the luminance of a measurement target site of the skin. In Patent Document 4 below, multi-resolution analysis is performed on skin image data to obtain values corresponding to the physical glossiness of the skin and the apparent roughness of the skin surface, and the skin glossiness is evaluated. A technique has been proposed.

  In Patent Documents 5 and 6 below, techniques focusing on face images are proposed. Specifically, Patent Document 5 below proposes a method of creating a synthesized face image having a different stereoscopic effect in a face image of a predetermined shape, focusing on the face image. In the above-mentioned Patent Document 6, the texture of the standardized face image is subjected to principal component analysis to calculate a weight coefficient of a multi-order basis vector, and the global appearance impression is evaluated based on the weight coefficient on the lower order side. A method for evaluating a local appearance impression based on a higher-order side weighting factor has been proposed.

JP 2004-215991 A JP 2009-134372 A JP 2009-131336 A JP 2004-166801 A JP 2008-276405 A JP 2012-8617 A

  However, each method proposed in Patent Documents 1 to 4 described above uses partial skin information, and thus there is a possibility that the appearance texture of the entire face of each subject cannot be accurately quantified. This is because the texture of the appearance can be obtained by comprehensively evaluating various factors such as unevenness of the skin and uneven color. Visual evaluation of the texture of the entire face is considered to contribute to information obtained from both the entire face and partial skin. Further, Patent Documents 5 and 6 do not propose evaluation of facial texture such as transparency.

  The present invention has been made in view of the above points, and provides a technique for quantifying the texture of the entire face with high accuracy based on a face image.

  Each aspect of the present invention employs the following configurations in order to solve the above-described problems.

  The face image analysis apparatus according to the first aspect obtains average face image information of a plurality of sample face images whose shapes are standardized and a plurality of basis functions obtained by principal component analysis on the plurality of sample face images. Sample acquisition means, principal component analysis means for calculating a basis coefficient of each basis function for the analysis target face image by principal component analysis of the analysis target face image whose shape is normalized, and a principal component Based on each basis coefficient calculated by the analysis unit and a plurality of basis functions and average face image information acquired by the sample acquisition unit, each reproduction face image corresponding to each of the plurality of cumulative contribution ratios of a predetermined accuracy is generated. The difference between the light and dark information between the lower-order reproduced face image and the higher-order reproduced face image generated by the reconstructing means with respect to the pair of adjacent cumulative contribution ratios. Thus, at least one of the bright difference image from which the darker component is removed than the lower-order reproduced face image and the dark difference image from which the brighter component is removed from the lower-order reproduced face image is used as the adjacent cumulative contribution rate. Difference calculation means for generating each pair, and at least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component from each color information in the light difference image and the dark difference image Representative value calculating means for calculating each of the adjacent cumulative contribution rate pairs, and at least one of the light component representative value and the dark component representative value for each pair of the adjacent cumulative contribution rates in ascending order of the cumulative contribution rate, respectively. Accumulating value calculating means for calculating at least one of a light component accumulated value and a dark component accumulated value for each pair of the adjacent accumulated contribution ratios by accumulation. That.

  In the face image analysis method according to the second aspect, at least one computer is obtained by average face image information of a plurality of sample face images whose shapes are standardized, and principal component analysis on the plurality of sample face images. By obtaining multiple basis functions, normalizing the shape of the analysis target face image, and performing principal component analysis on the analysis target face image whose shape is normalized, the basis coefficients of each basis function for the analysis target face image Respectively, and based on each calculated basis coefficient and the obtained plurality of basis functions and average face image information, each reproduced face image corresponding to each of the plurality of cumulative contribution rates of a predetermined accuracy is generated, By calculating the difference in light and dark information between the lower-order reproduced face image and the higher-order reproduced face image generated for the adjacent cumulative contribution rate pair, a darker component than the lower-order reproduced face image is obtained. At least one of the left light difference image and the dark difference image from which a brighter component than the lower-order reproduced face image is removed is generated for each pair of adjacent cumulative contribution ratios, and the light difference image and the dark difference image are generated. At least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component is calculated for each pair of the adjacent cumulative contribution rates from each color information in the difference image, and the adjacent By accumulating at least one of the light component representative value and the dark component representative value for each pair of cumulative contribution ratios in ascending order of the cumulative contribution ratio, the light component cumulative value and the dark component for each pair of adjacent cumulative contribution ratios. Calculating at least one of the cumulative values.

  As another aspect of the present invention, a program that causes a computer to realize each configuration according to the first aspect may be used, or a computer-readable recording medium that records such a program may be used. . This recording medium includes a non-transitory tangible medium.

  According to each aspect described above, it is possible to provide a technique for quantifying the texture of the entire face with high accuracy based on the face image. Note that quantification with high accuracy means obtaining a value that objectively represents the contents approximate to the sensory evaluation of a person, preferably a specialized panel, compared to the conventional case.

It is a figure which shows notionally the linear sum of an average face image and each eigenface image obtained by principal component analysis. It is a figure which shows notionally the example of production | generation of each reproduction face image. It is a figure which shows notionally the example of a production | generation of a bright difference image and a dark difference image. It is a figure which shows notionally the example of calculation of a bright component cumulative value. It is a figure which shows notionally the hardware structural example of the face image analyzer in 1st Embodiment. It is a figure which shows notionally the process structural example of the face image analyzer in 1st Embodiment. It is a figure which shows the example of the bright component graph regarding an analysis object face image. It is a figure which shows the example of the dark component graph regarding an analysis object face image. It is a figure which shows the example of the bright component graph regarding a reference | standard face image and an analysis object face image. It is a figure which shows the example of the dark component graph regarding a reference | standard face image and an analysis object face image. It is a flowchart which shows the operation example of the face image analyzer in 1st Embodiment. It is a figure which shows notionally the process structural example of the face image analyzer in 2nd Embodiment. It is a figure which shows the example of a bright component difference graph. It is a figure which shows the example of a dark component difference graph. It is a figure which shows the example of a comparison graph. It is a figure which shows the example of the comparison graph which uses a face image before makeup and a face image after makeup as an analysis object face image. It is a figure which shows the bright component graph regarding five transparent face-sensitive images. It is a figure which shows the dark component graph regarding five transparent face-sensitive images. It is a figure which shows the light component graph in the range of 68% or more and 98% or less of the accumulation contribution rate. It is a figure which shows the dark component graph in the range of 68 to 98% of the cumulative contribution rate. It is a figure which shows the light component graph regarding a typical transparent face image and a comparison object face image. It is a figure which shows the dark component graph regarding a typical transparent face image and a comparison object face image. It is a figure which shows the 16 face images corresponding to the combination of distance which each distance increases by the substantially equal width | variety, and the distance from the origin calculated regarding each face image. It is a figure which shows the difference of the bright component cumulative value regarding each face image shown by FIG. 17, the difference of dark component cumulative value, and the sensory evaluation result regarding each face image. It is a figure which shows the distance from the origin calculated regarding each face image shown in FIG. 17, and the sensory evaluation result regarding each face image. It is a figure which shows the example of the information which shows the texture analysis result for every range of the distance from an origin.

  Embodiments of the present invention will be described below. In addition, embodiment mentioned below is an illustration and this invention is not limited to the structure of the following embodiment.

  The face image analysis apparatus according to the present embodiment acquires average face image information of a plurality of sample face images whose shapes are standardized, and a plurality of basis functions obtained by principal component analysis on the plurality of sample face images. A sample acquisition unit, a principal component analysis unit that calculates a basis coefficient of each basis function for the analysis target face image by performing principal component analysis on the analysis target face image whose shape is normalized, and a principal component analysis Each reproduction face image corresponding to each of the plurality of cumulative contribution ratios of predetermined accuracy is generated based on each basis coefficient calculated by the unit and a plurality of basis functions and average face image information acquired by the sample acquisition unit. By calculating the difference in light and dark information between the reconstruction unit and the lower-order reproduced face image generated by the reconstruction unit with respect to the adjacent cumulative contribution rate pair, At least one of a bright difference image from which darker components are removed than the side reproduction face image and a dark difference image from which brighter components are removed than the lower-order reproduction face image is used for each of the adjacent cumulative contribution ratio pairs. The difference calculation unit to be generated, and at least one of the light component representative value indicating the intensity of the light component and the dark component representative value indicating the intensity of the dark component from the color information in the light difference image and the dark difference image are adjacent to each other. A representative value calculation unit for calculating each pair of cumulative contribution rates, and by accumulating at least one of the light component representative value and the dark component representative value for each pair of adjacent cumulative contribution rates in ascending order of the cumulative contribution rate. And an accumulated value calculating unit for calculating at least one of a light component accumulated value and a dark component accumulated value for each pair of the adjacent accumulated contribution rates.

  In the face image analysis method according to the present embodiment, at least one computer has a plurality of average face image information of a plurality of sample face images whose shapes are normalized and a plurality of principal component analyzes obtained from the plurality of sample face images. The basis function of each analysis function is obtained by normalizing the shape of the analysis target face image and performing principal component analysis on the analysis target face image whose shape is normalized. Respectively calculated, and based on each calculated basis coefficient and a plurality of obtained basis functions and average face image information, each reproduced face image corresponding to each of the plurality of cumulative contribution ratios of a predetermined accuracy is generated, By calculating the difference in light and dark information between the lower-order reproduced face image and the higher-order reproduced face image generated for adjacent cumulative contribution ratio pairs, a darker component than the lower-order reproduced face image is obtained. At least one of the left light difference image and the dark difference image from which a brighter component than the lower-order reproduced face image is removed is generated for each pair of adjacent cumulative contribution ratios, and the light difference image and the dark difference image are generated. At least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component is calculated for each pair of the adjacent cumulative contribution rates from each color information in the difference image, and the adjacent By accumulating at least one of the light component representative value and the dark component representative value for each pair of cumulative contribution ratios in ascending order of the cumulative contribution ratio, the light component cumulative value and the dark component for each pair of adjacent cumulative contribution ratios. Calculating at least one of the cumulative values.

Here, a plurality of sample face images whose shapes are standardized include an average face shape (shape of hairline, contour, eyebrows, eyes, nose, etc.) obtained from a plurality of sample face images, and a plurality of samples. It is formed from texture information unique to each face. The analysis target face image whose shape is standardized is formed from an average face shape obtained from the plurality of sample face images and texture information unique to the face of the subject. For normalization, for example, morphing processing such as AFIM (Automatic Facial Image Manipulation system) (see Reference 1 below) is used.
Reference 1: Takuda Terada et al., “Statistical analysis of facial textures using the automatic facial morphing system AFIM”, Journal of the Japanese Facial Society, October 5, 2009

  According to principal component analysis (PCA (Principal Component Analysis)) of a plurality of sample face images whose shapes are standardized, an N-dimensional eigenvector corresponding to the number of samples (N is the number of pixels of the face image) and each N-dimension The eigenvector weight coefficient is obtained. An N-dimensional eigenvector is expressed as a plurality of basis functions, and each weight coefficient is expressed as a basis coefficient. Then, according to the principal component analysis of the analysis target face image whose shape is normalized, the analysis target face image whose shape is normalized includes the average face image information of a plurality of sample face images whose shapes are normalized It is decomposed into a linear sum with a plurality of basis functions, and each basis coefficient corresponding to the face image to be analyzed is calculated.

The above equation shows the principal component analysis method. x is an input image, m is an average face, u ₁ , u ₂ ,. . . u _n is a basis function (eigenvector) arranged in descending order of eigenvalues. b represents a base coefficient, and v _R , v _G , and v _B represent the values of the pixels R, G, and B, respectively. x corresponds to a plurality of sample face images with normalized shapes and analysis target face images with normalized shapes, and m corresponds to the average face image information.

In the linear sum obtained by the principal component analysis, each basis function u _i is arranged in descending order of the eigenvalue calculated corresponding to each basis function. As shown in FIG. 1, the lower-order base represents a global texture tendency as when the analysis target face is viewed from a distance, and the higher-order base displays the analysis target face in near vision. It represents the tendency of local texture as if Note that the face image represented by each basis function is also referred to as a unique face image. FIG. 1 is a diagram conceptually showing a linear sum of an average face image and each unique face image obtained by principal component analysis. In each drawing including the face image thereafter, the periphery of the eyes is painted black from the viewpoint of personal information protection. This is not related to the processing of this embodiment.

  In the present embodiment, using the information obtained by principal component analysis, each reproduced face image corresponding to each cumulative contribution rate with a predetermined accuracy is generated for the analysis target face image. The predetermined accuracy means a step size of the cumulative contribution rate for generating the reproduced face image.

  Each basis function obtained by principal component analysis has a larger amount of information expressed as the corresponding eigenvalue increases. That is, the lower the basis, the greater the contribution rate to the input image. On the other hand, even if the bases have the same base number, the contribution ratios indicated by the bases may differ depending on the texture of the analysis target face image serving as the input image. Therefore, when each reproduced face image is generated corresponding to each base number, the degree of reproduction of each reproduced face image may be different for each analysis target face image serving as an input image.

  Therefore, in the present embodiment, the cumulative contribution rate (ACR (Accumulated Cover Rate)) as described above is used as an index unified for each face image to be analyzed. The cumulative contribution rate can be obtained by using each eigenvalue corresponding to each basis function, and indicates the degree of reproduction of the reproduced face image obtained by the linear sum from the average face image to a certain basis with respect to the input face image. That is, in the present embodiment, each reproduced face image is generated by a linear sum from an average face image to a base indicating a certain cumulative contribution rate.

  FIG. 2 is a diagram conceptually illustrating an example of generating each reproduced face image. In the example of FIG. 2, the predetermined accuracy (step size) of the cumulative contribution rate is set to 3%. A cumulative contribution of 50% is achieved with only the average face image, and a cumulative contribution of 100% is achieved with a linear sum of all bases.

  In this embodiment, using each reproduced face image generated in this manner, at least one of a bright difference image and a dark difference image (only a bright difference image, only a dark difference image, Any one of them) is generated respectively. As described above, each basis function represents a local texture tendency from a global texture tendency in ascending order of basis numbers. Therefore, according to the present embodiment, the difference between two reproduced face images corresponding to two adjacent cumulative contribution ratios is taken to obtain a detailed level with different influences on the texture of the entire face (from global tendency to local tendency). Each texture information can be obtained separately. Roughly speaking, it is possible to distinguish and acquire texture information that forms a global tendency with a low level of detail and texture information that forms a local tendency with a high level of detail.

  Further, since the light difference image and the dark difference image are calculated based on the difference in the light / dark information between the reproduced face images, according to the present embodiment, the light / darkness that easily affects the appearance as a psychological action associated with human vision. Characteristic information can be obtained. That is, by calculating the difference between the light and dark information, a shine component is extracted as a component that is viewed brighter than the average, and a color unevenness component and an uneven unevenness component are extracted as components that are viewed darker than the average. That is, according to the bright difference image, the shine component can be extracted, and according to the dark difference image, the color unevenness component and the uneven unevenness component can be extracted. As the brightness information, for example, brightness information, brightness information, RGB G value and the like can be used, but it is desirable to use brightness information. Color unevenness components are dirt, stains, freckles, and the like of pores, and uneven unevenness components are pores and wrinkles.

  FIG. 3 is a diagram conceptually illustrating an example of generating a bright difference image and a dark difference image. In the example of FIG. 3, the predetermined accuracy of the cumulative contribution rate is set to 3%, and a bright difference image is obtained from a difference between two reproduced face images corresponding to two cumulative contribution rates adjacent to each other with a step size of 3%. And a pair of dark difference images are generated.

  In the present embodiment, in order to digitize the texture information acquired as described above, at least one of the bright component representative value and the dark component representative value is calculated for the bright difference image and the dark difference image, and further, the accumulation of these values is calculated. As a value, at least one of a bright component cumulative value and a dark component cumulative value is calculated for each pair of adjacent cumulative contribution rates. As described above, the bright difference image includes the shine component, and the dark difference image includes the color unevenness component and the uneven unevenness component. Therefore, the bright component representative value obtained from the bright difference image indicates the strength of the shine component. The dark component representative value obtained from the dark difference image indicates the strength of the color unevenness component and the unevenness unevenness component. The representative values here are, for example, histogram average values, variance values, standard deviations and other statistical values, and histogram areas. With respect to the representative value, the specific form is not limited as long as the magnitude of the bright component and the dark component is indicated in some form.

  FIG. 4 is a diagram conceptually illustrating a calculation example of the bright component cumulative value. In the graph of FIG. 4, each light component accumulated value calculated for each pair of adjacent accumulated contribution rates is plotted on the higher accumulated contribution rate in the pair. According to the present embodiment, a graph like the example of FIG. 4 can be generated from at least one of the bright component cumulative value and the dark component cumulative value calculated for each pair of adjacent cumulative contribution rates. Note that the dark component cumulative value is also calculated in the same manner as in FIG.

  As described above, according to the present embodiment, the facial texture information having different influences on the texture of the entire face of the analysis target facial image is obtained by the image processing in which the facial texture characteristics and the psychological effects associated with human vision are reproduced. For each detail level, at least one of an index relating to the shine component (representative value of the bright component) and an index relating to the color unevenness component and the unevenness unevenness component (representative value of the dark component) can be accurately acquired. Furthermore, according to the present embodiment, the cumulative values of at least one of the index relating to the shine component and the index relating to the color unevenness component and the uneven unevenness component are calculated in ascending order of the detail levels having different effects on the texture of the entire face can do.

  For example, according to FIG. 4, since the horizontal axis corresponds to the above detailed level and the vertical axis corresponds to the index relating to the shine component at the detailed level, The shine condition that affects the texture of the face can be evaluated. Similarly, it is possible to evaluate the color unevenness state and the uneven unevenness state with respect to the face to be analyzed from the shape of the dark component cumulative value locus.

  As described above, according to the present embodiment, the texture of the entire face in the face image to be analyzed is determined by the cumulative value selected from at least one of the bright component cumulative value and the dark component cumulative value for each pair of adjacent cumulative contribution rates. Can be quantified with high accuracy. By using both the light component cumulative value and the dark component cumulative value, it is possible to obtain both an index relating to the shine component and an index relating to the color unevenness component and the uneven unevenness component. The texture of the entire face can be quantified with higher accuracy. Here, the texture of the entire face is the texture of the entire face selected from a feeling of skin, transparency, gloss, darkness, and unevenness when a certain person views the entire face of the subject. Is transparent. It is a feeling that the person will receive from textures other than the shape of the subject's face.

Hereinafter, the details of the above-described embodiment will be described.
[First Embodiment]

〔Device configuration〕
FIG. 5 is a diagram conceptually illustrating a hardware configuration example of the face image analysis apparatus (hereinafter also referred to as an analysis apparatus) 10 according to the first embodiment. The analysis apparatus 10 according to the first embodiment is a so-called computer, and includes, for example, a CPU (Central Processing Unit) 2, a memory 3, an input / output interface (I / F) 4, and the like that are connected to each other via a bus 5. The memory 3 is a RAM (Random Access Memory), a ROM (Read Only Memory), a hard disk, a portable storage medium, or the like.

  The input / output I / F 4 is connected to a communication device or the like that communicates with other computers via the input unit 7, the output unit 9, and a network (not shown). The input unit 7 is a device that receives input of a user operation such as a keyboard and a mouse. The output unit 9 is a device that provides information to a user, such as a display device or a printer. Note that the hardware configuration of the analyzer 10 is not limited.

  FIG. 6 is a diagram conceptually illustrating a processing configuration example of the analysis apparatus 10 in the first embodiment. The analysis apparatus 10 according to the first embodiment includes an image acquisition unit 11, a normalization processing unit 12, a principal component analysis unit 13, a sample data storage unit 14, a reconstruction unit 15, a difference calculation unit 16, a representative value calculation unit 17, and a cumulative value. A value calculation unit 18, an information generation unit 19, a reference data storage unit 20, and the like are included. Each of these processing units is realized, for example, by executing a program stored in the memory 3 by the CPU 2. Further, the program may be installed from a portable recording medium such as a CD (Compact Disc) or a memory card or another computer on the network via the input / output I / F 4 and stored in the memory 3. Good.

  The image acquisition unit 11 acquires image data of an analysis target face image in which the face of the subject is captured. This image data is acquired from a portable recording medium, another computer or the like via the input / output I / F 4. In the present embodiment, the data format of the image data is not limited. Therefore, the image data is acquired as a file in a JPEG (Joint Photographic Experts Group) format, a GIF (Graphic Interchange Format) format, or the like. Hereinafter, the image data of the face image is also simply referred to as a face image.

  The face image acquired by the image acquisition unit 11 only needs to include the entire face of the subject and may include parts other than the background and the head. The subject's face may be a natural face or a face with makeup (makeup face). Further, the face image may be a color image or a gray scale image as long as it holds brightness / darkness information.

  The normalization processing unit 12 performs normalization processing such as normalization on the analysis target face image acquired by the image acquisition unit 11 to generate a standardized image in which the shape characteristics of the subject's face are discarded. That is, the normalization processing unit 12 generates an analysis target face image whose shape is standardized as described above. An analysis target face image whose shape is standardized and handled by the normalization processing unit 12 and later is simply referred to as an analysis target face image.

  The sample data storage unit 14 stores average face image information and a plurality of basis functions acquired by the sample acquisition unit of the above-described embodiment. The average face image information and the basis function are as described above.

  The principal component analysis unit 13 performs principal component analysis on the analysis target face image processed by the normalization processing unit 12 using the average face image information stored in the sample data storage unit 14 and the plurality of basis functions. A basis coefficient of each basis function for the analysis target face image is calculated. Since the principal component analysis unit 13 corresponds to the principal component analysis unit of the above-described embodiment, the details are as described above.

  Based on each basis coefficient calculated by the principal component analysis unit 13 and a plurality of basis functions and average face image information stored in the sample data storage unit 14, the reconstruction unit 15 has a predetermined lower limit value greater than 50%. Each reproduction face image corresponding to each cumulative contribution rate with a predetermined accuracy in a range from 1 to a predetermined upper limit value less than 100% is generated. The predetermined lower limit value, the predetermined upper limit value, and the predetermined accuracy are determined in advance and stored in the memory 3 from the following viewpoints.

  The inventors of the present invention have a texture contribution range that represents texture information that has a particularly large influence on the texture of the entire face and a range that represents other texture information within the entire range of the cumulative contribution ratio of 50% to 100%. It has been clarified that the texture contribution range is 50% or more and less than the predetermined lower limit value and does not include the predetermined upper limit value or more and 100% or less. This is a new technical idea that is different from the common sense of the related art of focusing on low-order terms, that is, global texture trends, in order to be able to determine the overall impression of the face. They have found that texture information that contributes to the texture of the entire face exists in a range of cumulative contributions that does not necessarily match the range of cumulative contributions that can determine the overall impression of the face. In the present embodiment, the predetermined lower limit value and the predetermined upper limit value are set in advance so as to indicate a texture contribution range representing texture information that has a particularly large influence on the texture of the entire face. In addition, the consideration about the specific numerical value of the said predetermined | prescribed lower limit value and the said predetermined | prescribed upper limit value for setting a texture contribution range is demonstrated in the item of an Example.

  On the other hand, if the predetermined accuracy (step size) related to the cumulative contribution rate is too fine, there is no difference in light / dark information for each pair of adjacent cumulative contribution rates, and texture information to be obtained from the light difference image and the dark difference image May not be available. On the other hand, if the predetermined accuracy of the cumulative contribution rate is too coarse, the difference in the average value of the light / dark information between the adjacent reproduced face images of the cumulative contribution rate becomes large. In some cases, information on the deviation of the average value of the brightness information is included, and the texture information desired to be acquired cannot be acquired accurately. Therefore, the predetermined accuracy is the difference in the brightness information for each pair of adjacent cumulative contribution ratios according to the quality of the sample face image for obtaining a plurality of basis functions stored in the sample data storage unit 14 and the number of samples. Is set in advance to a value that appropriately matches the difference between the average values of the light and dark information between the reproduced face images of adjacent cumulative contribution ratios, and is stored in the memory 3. A discussion of specific numerical values with a predetermined accuracy regarding the cumulative contribution rate will also be described in the section of the embodiment.

  The difference calculation unit 16 calculates a difference in light and dark information between the low-order side reproduction face image and the high-order side reproduction face image generated by the reconstruction unit 15 with respect to adjacent cumulative contribution rate pairs. A bright difference image from which darker components are removed than the reproduced face image and a dark difference image from which brighter components are removed from the lower-order reproduced face image are generated for each pair of adjacent cumulative contribution rates.

  The representative value calculation unit 17 obtains a bright component representative value indicating the intensity of the bright component and a dark component representative value indicating the intensity of the dark component from each color information in the bright difference image and the dark difference image generated by the difference calculation unit 16. , Each pair of adjacent cumulative contribution rates is calculated.

  Based on the light component representative value and the dark component representative value calculated by the representative value calculation unit 17, the cumulative value calculation unit 18 accumulates the light component representative value and the dark component representative value for each pair of adjacent cumulative contribution ratios. Accumulated in ascending order of contribution rate, and the light component cumulative value and dark component cumulative value are calculated for each pair of adjacent cumulative contribution rates. The bright component cumulative value and dark component cumulative value calculated here are cumulative values within the range of the cumulative contribution rate from the predetermined lower limit value to the predetermined upper limit value.

  The reference data storage unit 20 stores a group of bright component cumulative values and dark component cumulative values for each pair of adjacent cumulative contribution rates corresponding to the reference face image to be compared with the analysis target face image. The group of the light component accumulated value and the dark component accumulated value corresponding to the reference face image stored in the reference data storage unit 20 is acquired from the portable recording medium, another computer, or the like via the input / output I / F 4. Alternatively, the reference face image instead of the analysis target face image may be acquired by being processed by the above-described processing units.

  The reference face image only needs to include the entire face according to the comparison viewpoint regarding the texture of the face. For example, the face image includes the entire face determined to be transparent by sensory evaluation by a professional evaluator or the like. Is selected as the reference face image. Further, the reference face image may be a color image or a gray scale image as long as it holds light and dark information.

  The information generation unit 19 forms the following bright component graph and dark component graph using the bright component cumulative value and the dark component cumulative value for each pair of adjacent cumulative contribution ratios calculated by the cumulative value calculation unit 18. Data to generate. In the bright component graph, the coordinates including the first axis indicating the cumulative contribution rate and the second axis indicating the bright component cumulative value are calculated for each pair of the adjacent cumulative contribution rates calculated by the cumulative value calculation unit 18. The light component cumulative value is plotted. In the dark component graph, the coordinates including the first axis indicating the cumulative contribution rate and the second axis indicating the dark component cumulative value are calculated for each pair of adjacent cumulative contribution rates calculated by the cumulative value calculation unit 18. The dark component cumulative value is plotted. The data for forming the graph is data for outputting the graph to the output unit 9 and may be screen display data, print data, or screen display data. It may be a data file having data or print data.

  FIG. 7A is a diagram illustrating an example of a bright component graph related to an analysis target face image. FIG. 7B is a diagram illustrating an example of a dark component graph relating to an analysis target face image. In the example of FIGS. 7A and 7B, the range of the cumulative contribution rate to be accumulated determined by the predetermined lower limit value and the predetermined upper limit value is set to 68% to 98%. In the example of FIGS. 7A and 7B, the dispersion value is used for the light component representative value and the dark component representative value that are the basis of the light component graph and the dark component graph. Further, in the graphs shown in FIGS. 7A and 7B, each bright component cumulative value and each dark component cumulative value is plotted on the higher cumulative contribution rate in the adjacent cumulative contribution rate pair.

  Further, the information generation unit 19 extracts a group of bright component cumulative values and dark component cumulative values for each pair of adjacent cumulative contribution ratios related to the reference face image from the reference data storage unit 20, and extracts the extracted light components. Data for forming the bright component graph and the dark component graph may be generated so that a group of component cumulative values and dark component cumulative values is superimposed on those of the analysis target face image. In this case, the information generation unit 19 corresponds to a reference acquisition unit.

  FIG. 8A is a diagram illustrating an example of a bright component graph related to the reference face image and the analysis target face image. FIG. 8B is a diagram illustrating an example of a dark component graph regarding the reference face image and the analysis target face image. Also in the examples of FIGS. 8A and 8B, the range of the cumulative contribution rate to be accumulated determined by the predetermined lower limit value and the predetermined upper limit value is set to 68% to 98%. In the example of FIGS. 8A and 8B, the dispersion value is used for the light component representative value and the dark component representative value that are the basis of the light component graph and the dark component graph. In the light component graph in this example, the light component cumulative value for each pair of adjacent cumulative contribution ratios related to the analysis target face image is superimposed on the reference face image. Similarly, in the dark component graph, for the analysis target face image, that for the reference face image is superimposed on the dark component cumulative value for each pair of adjacent cumulative contribution ratios.

  The information generation unit 19 may generate data for forming the above-described four types of all graphs, or for forming any one or any of the plurality of graphs. Data may be generated. 7A, 7B, 8A, and 8B, the first axis is set on the horizontal axis and the second axis is set on the vertical axis. The dark component graph is not limited to such an example.

[Operation example]
Hereinafter, the face image analysis method according to the first embodiment will be described with reference to FIG. FIG. 9 is a flowchart illustrating an operation example of the analyzer 10 according to the first embodiment. In the example of FIG. 9, an example of generating data for forming a light component graph and a dark component graph (see the examples of FIGS. 8A and 8B) in which data related to the reference face image and the analysis target face image are superimposed is generated. Are listed. Also, at this time, the analysis apparatus 10 already has a group of bright component cumulative values and dark component cumulative values for each pair of adjacent cumulative contribution ratios corresponding to the reference face image for comparison with the analysis target face image. Assume that it is retained.

The analysis apparatus 10 acquires an analysis target face image in which the face of the subject is captured (S91).
Subsequently, the analysis apparatus 10 performs normalization processing such as normalization on the analysis target face image (S92). As a result, an analysis target face image whose shape is standardized is generated. In the subsequent processes, the analysis target face image whose shape is standardized is handled, and thereafter, the analysis target face image whose shape is standardized is simplified and described as the analysis target face image.

  The analysis apparatus 10 performs principal component analysis on the analysis target face image generated in the step (S92) (S93). In this principal component analysis, the analysis apparatus 10 uses a plurality of basis functions acquired in advance by principal component analysis of a plurality of sample face images whose shapes are normalized, and uses each basis function for the analysis target face image. Each basis coefficient is calculated. That is, the analysis target face image whose shape is normalized is decomposed into a linear sum of the average face image information of the plurality of sample face images whose shapes are normalized and the plurality of basis functions, and the analysis target face image is converted into the analysis target face image. Each corresponding basis coefficient is calculated.

  Subsequently, the analysis apparatus 10 based on each basis coefficient calculated in the step (S93) and a plurality of basis functions and average face image information calculated in advance by principal component analysis of a plurality of sample face images. Reproduced face images corresponding to each cumulative contribution rate with a predetermined accuracy in a range from a predetermined lower limit value greater than 50% to a predetermined upper limit value less than 100% are generated (S94).

  The analysis apparatus 10 calculates a difference in light and dark information between the lower-order reproduced face image and the higher-order reproduced face image generated in the step (S94) with respect to adjacent cumulative contribution rate pairs. Thereby, the analysis apparatus 10 adjoins the bright difference image from which the darker component than the low-order side reproduction face image is removed, and the dark difference image from which the brighter component than the low-order side reproduction face image is removed. Each pair of cumulative contribution rates is generated (S95).

  The analysis apparatus 10 calculates, from each color information in the bright difference image and the dark difference image generated in the step (S95), a bright component representative value indicating the intensity of the bright component and a dark component representative value indicating the intensity of the dark component. Each pair of adjacent cumulative contribution rates is calculated (S96).

  Subsequently, the analysis apparatus 10 accumulates the bright component representative value and the dark component representative value for each pair of the adjacent cumulative contribution ratios calculated in the step (S96) in ascending order of the cumulative contribution ratios, and the adjacent The bright component cumulative value and the dark component cumulative value are calculated for each pair of cumulative contribution ratios to be performed (S97).

  The analysis apparatus 10 includes a bright component cumulative value and a dark component cumulative value for each pair of adjacent cumulative contribution ratios related to the reference face image, and a light component for each pair of adjacent cumulative contribution ratios related to the analysis target face image. Data for forming a light component graph and a dark component graph on which the component cumulative value and the dark component cumulative value are superimposed is generated (S98). Thereafter, based on this data, graphs such as the examples of FIGS. 7A, 7B, 8A, and 8B are output.

  Here, with respect to the reference face image, the group of the bright component cumulative value and the dark component cumulative value for each pair of the adjacent cumulative contribution ratios is determined from the above-described steps (S91) to (S97). Instead, it may be acquired by being executed on the reference face image.

[Operation and Effect of First Embodiment]
As described above, in the first embodiment, the light component cumulative value and the dark component cumulative value are calculated for each pair of adjacent cumulative contribution rates in the cumulative contribution rate range from the predetermined lower limit value to the predetermined upper limit value. . As described above, the bright component cumulative value for a certain pair of adjacent cumulative contribution ratios is a light component between two reproduced face images corresponding to each pair of adjacent cumulative contribution ratios among the cumulative contribution ratios lower than that pair. Each difference of component information is a value accumulated in ascending order of the cumulative contribution rate. Similarly, as described above, the dark component cumulative value for a certain pair of adjacent cumulative contribution ratios is two reproduced face images corresponding to each pair of adjacent cumulative contribution ratios among cumulative contribution ratios lower than that pair. Each difference in dark component information is a value accumulated in ascending order of the cumulative contribution rate. Furthermore, the range of the cumulative contribution rate in which the bright component cumulative value and the dark component cumulative value are calculated is set to a range representing texture information that has a particularly large influence on the texture of the entire face.

  Thus, the bright component cumulative value indicates the strength of the shine component that is included in the reproduced face image of the cumulative contribution ratio of the corresponding pair and has a large influence on the texture of the entire face, and the dark component cumulative value indicates the corresponding pair The intensity of the color unevenness component and the unevenness unevenness component having a large influence on the texture of the entire face, included in the reproduced face image of the cumulative contribution rate of. In the first embodiment, data for forming a light component graph and a dark component graph, in which a light component cumulative value and a dark component cumulative value are plotted, is generated for each cumulative contribution rate with a predetermined accuracy.

  Therefore, according to the first embodiment, the light component graph and the dark component graph can objectively quantify the strength of the shine component, the color unevenness component, and the uneven unevenness component that have a large influence on the texture of the entire face. As a result, the texture of the entire face of the subject can be quantified with high accuracy.

  In addition, according to the first embodiment, each bright component cumulative value is represented for each pair of adjacent cumulative contribution ratios, so that each level of detail (from global trend to local trend) has a different influence on the texture of the entire face. The strength of the shine component can be distinguished and acquired. Similarly, by expressing each dark component cumulative value for each pair of adjacent cumulative contribution ratios, color unevenness components and uneven unevenness components for each level of detail (global trend to local trend) that have different effects on the texture of the entire face. Can be obtained by distinguishing the strength of. For example, according to the example of FIG. 7A, from the cumulative contribution rate pair including the predetermined lower limit value to the cumulative contribution rate pair including the predetermined upper limit value, that is, from the global texture to the local texture, the bright component cumulative value However, the rate of increase is almost uniform. In this case, it can be analyzed that there is a shine component that affects the texture of the entire face both globally and locally. On the other hand, according to the example of FIG. 7B, the increase from the dark component cumulative value plotted on the cumulative contribution rate of 74% to the dark component cumulative value plotted on the cumulative contribution rate of 80% It is larger than the extent of the range. In this case, it can be analyzed that at least one of the color unevenness component and the unevenness unevenness component is present in the texture represented by the range of the cumulative contribution ratio having a large increase width.

  Furthermore, in the first embodiment, the light component graph in which the light component accumulated value and the dark component accumulated value for each pair of the adjacent accumulated contribution ratios related to the reference face image are superimposed on the data related to the analysis target face image described above. And data for forming a dark component graph is generated. Therefore, according to the first embodiment, by using a face image determined to be beautiful skin or transparent skin by sensory evaluation as the reference face image, it is generated from the entire face of the reference face and the face of the subject. The difference in texture can be shown objectively. For example, according to the examples of FIGS. 8A and 8B, it can be seen that there is a certain difference in the texture of the entire face between the face of the subject to be analyzed and the reference face. Furthermore, for example, by comparing the light component graph of FIG. 8A and the dark component graph of FIG. 8B, the texture of the entire face of the subject is more largely derived from at least one of the color unevenness component and the uneven unevenness component than the glossy component. You can also analyze it.

[Second Embodiment]
In the second embodiment, the difference between the light component accumulated value and the dark component accumulated value between the reference face image and the analysis target face image is used as an index for analyzing the face image. Hereinafter, the analyzer 10 according to the second embodiment will be described focusing on the content different from the first embodiment. In the following description, the same contents as those in the first embodiment are omitted as appropriate.

  FIG. 10 is a diagram conceptually illustrating a processing configuration example of the analysis apparatus 10 in the second embodiment. The analysis apparatus 10 in the second embodiment further includes a comparison unit 21 and an analysis information generation unit 22 in addition to the configuration of the first embodiment. The comparison unit 21 and the analysis information generation unit 22 are realized by executing a program stored in the memory 3 by the CPU 2, for example, as with the other processing units.

  The comparison unit 21 extracts a group of light component accumulated values and dark component accumulated values related to the reference face image from the reference data storage unit 20, and is extracted from the reference data storage unit 20 for each pair of adjacent accumulated contribution ratios. The difference between the light component accumulated value and the light component accumulated value calculated by the accumulated value calculating unit 18, the dark component accumulated value acquired from the reference data storage unit 20, and the accumulated value calculating unit 18 The difference from the dark component cumulative value is calculated.

  The information generation unit 19 calculates the adjacent cumulative contribution rate calculated by the comparison unit 21 on the coordinates including the first axis indicating the cumulative contribution rate and the second axis indicating the difference calculated by the comparison unit 21. The light component difference graph in which the difference in the light component accumulated value for each pair is plotted, and the difference in the dark component accumulated value for each pair of the adjacent accumulated contribution ratio calculated by the comparison unit 21 is plotted. Data for forming at least one of the dark component difference graphs is generated.

  FIG. 11A is a diagram illustrating an example of a bright component difference graph. FIG. 11B is a diagram illustrating an example of a dark component difference graph. In the example of FIGS. 11A and 11B, the range of the cumulative contribution rate to be accumulated determined by the predetermined lower limit value and the predetermined upper limit value is set to 68% to 98%. Further, in the examples of FIGS. 11A and 11B, variance values are used for the light component representative value and the dark component representative value that are the basis of the light component graph and the dark component graph. 11A and 11B, since the analysis target face image B has a small difference between adjacent pairs of cumulative contribution ratios, the texture of the entire face that appears in the analysis target face image B appears in the reference face image. It can be determined that it is close to that of the face. On the other hand, since the analysis target face image A has a difference in both the light component difference graph and the dark component difference graph, the face reflected in the analysis target face image A is a shine component rather than the face reflected in the reference face image. It can be determined that there are many color unevenness components and unevenness unevenness components. Furthermore, in the pair including the upper limit value of the cumulative contribution rate, since the difference is larger in the dark component difference graph than in the light component difference graph, the face appearing in the analysis target face image A is more uneven in color and unevenness than the shine component. It can also be determined that there are more uneven components.

  The information generation unit 19 is calculated by the comparison unit 21 at coordinates including a first axis indicating a difference regarding the bright component cumulative value calculated by the comparison unit 21 and a second axis indicating a difference regarding the dark component cumulative value. You may make it produce | generate the data for forming the comparison graph by which the difference regarding the light component accumulation value and dark component accumulation value about each pair of the said adjacent accumulation contribution rate was plotted.

  FIG. 12 is a diagram illustrating an example of a comparison graph. In the comparison graph shown in FIG. 12, the horizontal axis indicates the difference regarding the bright component cumulative value, and the vertical axis indicates the difference regarding the dark component cumulative value. Each plot shows a corresponding pair of adjacent cumulative contributions. As shown in FIG. 12, according to the comparison graph, when the texture of the entire face is close to the face reflected in the reference face image, plots gather near the origin, and the texture of the entire face is obtained from the face reflected in the reference face image. If far away, the plot is placed away from the origin.

  The analysis information generation unit 22 includes the light component accumulation value and the dark component accumulation for any one pair of adjacent accumulation contribution ratios among the differences regarding the light component accumulation value and the dark component accumulation value calculated by the comparison unit 21. The distance from the origin of the position indicated by each difference regarding the value is calculated, and based on the calculated distance, the texture analysis information of the face appearing in the analysis target face image is generated based on the face appearing in the reference face image. . It is desirable that the distance calculation target be a difference between the bright component cumulative value and the dark component cumulative value calculated for adjacent pairs of cumulative contribution rates including the predetermined upper limit value. In the example of FIG. 12, each distance from the origin (0, 0) of each plot surrounded by a circle is calculated. The analysis information generation unit 22 holds information indicating the texture analysis result for each distance range in the memory 3 in advance, and generates the texture analysis information of the face that appears in the analysis target face image using this information.

[Operation and Effect of Second Embodiment]
In the second embodiment, for each pair of adjacent cumulative contribution ratios, the difference between the bright component cumulative value related to the reference face image and the bright component cumulative value related to the analysis target face image, and the dark component cumulative value related to the reference face image And the difference between the dark component accumulated value relating to the analysis target face image. As described above, according to the second embodiment, the difference in the strength of the shine component between the reference face and the analysis target face, and the difference in the strength of the color unevenness component and the unevenness unevenness component are determined from the global tendency. It is possible to quantify each detail level of the texture up to local tendency.

  Furthermore, in the second embodiment, a light component difference graph in which the difference in the light component accumulated value is plotted for each pair of adjacent accumulated contribution rates, and the difference in the dark component accumulated value for each pair of adjacent accumulated contribution rates Are output at least one of the dark component difference graphs plotted. According to the bright component difference graph, it is possible to easily grasp the difference regarding the shine component between the reference face and the analysis target face at each detailed level of the face texture.

  Furthermore, in the second embodiment, based on the distance from the origin of the position indicated by the difference between the light component accumulated value and the difference between the dark component accumulated values, the image appears in the analysis target face image based on the face in the reference face image. Face texture analysis information is generated. Thereby, according to 2nd Embodiment, objective texture-analysis information of a test subject's face can be automatically produced | generated by using a reference | standard face image and an analysis object face image. For example, when an image showing a face evaluated as having a beautiful skin feeling is used as a reference face image, analysis information (closeness to the beautiful skin, etc.) relating to the beautiful skin feeling of the subject's face is automatically generated.

[Supplement]
In each of the embodiments described above, the face captured in the analysis target face image is not particularly limited, and the overall texture of the face itself is quantified. However, depending on how the analysis target face image and the reference face image are selected. The makeup effect and the like can also be evaluated using the quantified information.

  For example, when the image acquisition unit 11 acquires a face image before makeup with a normalized shape and a face image after makeup with a normalized shape as the analysis target face image, the information generation unit 19 The data is generated so that the light component cumulative value and the dark component cumulative value calculated for the pre-makeup face image and the post-makeup face image are superimposed on the dark component graph, respectively. In addition, the information generation unit 19 superimposes the difference of the light component accumulated value and the difference of the dark component accumulated value respectively calculated on the face image before makeup and the face image after makeup on the light component difference graph and the dark component difference graph. As such, the data may be generated. In addition, the information generation unit 19 generates the data so that the difference between the bright component accumulated value and the dark component accumulated value calculated for the pre-makeup face image and the post-makeup face image are superimposed on the comparison graph. You may make it do. Note that the image acquisition unit 11 may acquire the pre-makeup face image as the reference face image and acquire the post-makeup face image as the analysis target face image.

  FIG. 13 is a diagram illustrating an example of a comparison graph in which the face image before makeup and the face image after makeup are analysis target face images. According to FIG. 13, since the distance from the origin corresponding to the post-makeup face image is shorter than the distance from the origin corresponding to the pre-makeup face image, the shine component, unevenness component and uneven unevenness component It can be determined that it has been sufficiently erased. In this case, the analysis information generation unit 22 calculates the distance from the origin for each of the post-makeup face image and the pre-makeup face image, and is applied to the makeup face reflected in the post-makeup face image based on the calculated difference in distance. Alternatively, makeup effect information indicating the effect of makeup may be generated as texture analysis information.

  As another example, the image acquisition unit 11 may acquire, as analysis target face images, a plurality of makeup face images whose shapes are standardized and that indicate changes in the makeup face over time. In this case, the texture of each makeup face image before and after the elapse of a predetermined time can be quantified using the graph output from the data generated by the information generation unit 19. Further, the distance from the origin can be calculated for each makeup face image before and after the predetermined time has elapsed. Thereby, the temporal collapse information of the makeup can be acquired as the texture analysis information.

  As another example, the image acquisition unit 11 may acquire a skin care pre-face image before using the skin care cosmetic and a skin care post-face image after using the skin care cosmetic as the analysis target face image. Good. According to this, the texture of the face before and after using the skin care cosmetic can be quantified, and the effect of the skin care cosmetic can also be evaluated.

[Modification]
In each of the embodiments described above, the generation target of the reproduced face image in the reconstruction unit 15 is limited to the range of the cumulative contribution rate from the predetermined lower limit value to the predetermined upper limit value. Each reproduced face image may be generated for the entire range of the cumulative contribution rate of 100% or less. In this case, the difference calculation unit 16, the representative value calculation unit 17, or the cumulative value calculation unit 18 finally has a bright component cumulative value corresponding to the cumulative contribution rate range from the predetermined lower limit value to the predetermined upper limit value. In addition, the reproduction face image to be calculated may be limited so that the cumulative dark component value is generated.

  Examples will be given below to describe the above-described embodiments in more detail. The present invention is not limited by the following examples. In the following example, 277 face images with unified viewpoint, illumination, and shape conditions are used as a plurality of sample face images whose shapes are standardized, and the predetermined accuracy of the cumulative contribution rate is 3%. Is set.

  In this example, each of the five face images (hereinafter referred to as “transparency face images”) that the expert evaluator felt transparent was used as the analysis target face image, and the above-described method of the first embodiment was performed. . FIG. 14A is a diagram illustrating a light component graph regarding five transparent face-sensitive images, and FIG. 14B is a diagram illustrating a dark component graph regarding five transparent face-sensitive images. Here, when the light component cumulative value and the dark component cumulative value are obtained by limiting to the range of the cumulative contribution ratio of 68% or more and 98% or less, graphs as shown in FIGS. 15A and 15B are obtained.

  FIG. 15A is a diagram showing a light component graph in the range of cumulative contribution ratios of 68% or more and 98% or less, and FIG. 15B is a diagram showing a dark component graph in the range of cumulative contribution ratios of 68% or more and 98% or less. is there. As shown in FIG. 15A and FIG. 15B, the light component cumulative value and the dark component cumulative value for each pair of adjacent cumulative contribution rates are in the range of 68% to 98% cumulative contribution rate. The same tendency is shown between images. Therefore, under the conditions of this embodiment (number of samples: 277, cumulative contribution rate accuracy: 3%), the texture contribution range having texture information that has a particularly large effect on the transparency of the entire face is in the range of 68% to 98%. Can be set to

  In the present embodiment, the range of 68% to 98% is set as the texture contribution range of the cumulative contribution rate, but the range in which the plot trajectory of each transparent face image shows a similar shape is the texture contribution range. May be set as. Therefore, according to the example of FIGS. 14A and 14B, the predetermined lower limit value indicating the texture contribution range may be set to 59% or more and less than 68%, and the predetermined upper limit value indicating the texture contribution range is set to about 95%. May be. Furthermore, the texture contribution range of the cumulative contribution rate may be set to the inside of the above-described range of 68% to 98%, that is, a range in which the predetermined upper limit value is 68% or more and the predetermined lower limit value is 98% or less.

  Here, the light component cumulative value and the dark component cumulative value in the five transparent face-sensitive images described above are averaged to calculate the light component cumulative value and the dark component cumulative value related to the representative transparent face-sensitive image, and the calculation The face image corresponding to the recorded data is positioned as a representative transparent face image. In this example, the relationship between this representative transparent face image and other face images was compared. As comparison objects, face images having a transparency obstruction factor such as a shine component, color unevenness component, and uneven unevenness component (indicated as “transparency inhibition face” in FIGS. 16A and 16B) and transparency obstruction factors are few. A face image of a face that can be felt transparent (noted as a beautiful skin face in FIGS. 16A and 16B) was used.

  FIG. 16A is a diagram illustrating a light component graph regarding a representative transparent face image and a comparison target face image, and FIG. 16B is a diagram illustrating a dark component graph regarding a typical transparent face image and a comparison target face image. is there. As shown in FIG. 16A and FIG. 16B, the light component cumulative value and the dark component cumulative value in the range of the cumulative contribution ratio of 68% or more and 98% or less are between a representative transparent face image and a beautiful skin face image. In the same way, the increase rate of the cumulative contribution rate tended to be small, and the tendency was different between the representative transparent face image and the transparency-inhibited face image. This comparison also proves that the texture contribution range of the cumulative contribution rate as described above has texture information that has a particularly large influence on the transparency of the entire face. In addition, in each of the above-described embodiments, it is also proved that the light component cumulative value and the dark component cumulative value calculated for each pair of adjacent cumulative contribution ratios quantify the texture of the entire face with high accuracy. Is done.

  Furthermore, in the present embodiment, the plurality of randomly selected face images are set as analysis target face images, the reference face image is set as the above-described representative transparent face image, and each analysis target face image is set as described in the second embodiment. In the embodiment, the distance from the origin of the position indicated by the difference between the bright component accumulated value and the dark component accumulated value was calculated. Then, a combination of distances (corresponding to 16 face images) is extracted from the calculated distances so that each distance increases with a substantially equal width, and the second implementation is performed for each face image corresponding to the combination. Each distance calculated by the method of form was compared with the result of the sensory evaluation (psychological evaluation) of the expert evaluator.

  FIG. 17 is a diagram illustrating 16 face images corresponding to combinations of distances in which the distances increase with substantially equal widths, and distances from the origin calculated for the face images. As shown in FIG. 17, according to the method of the second embodiment, it is possible to analyze that the face image B61 is farthest from the representative transparent face image, that is, has the largest transparency impeding factor. It is possible to analyze that the face image B16 is most similar to the representative transparent face image, that is, has the smallest transparency hindrance factor.

FIG. 18 is a diagram illustrating the difference between the bright component cumulative value and the dark component cumulative value regarding each face image shown in FIG. 17 and the sensory evaluation result regarding each face image. Each plot shown in FIG. 18 shows the difference between the light component cumulative value and the dark component cumulative value calculated for each of the 16 face images, and the numerical value given to each plot is the sensory value of the expert evaluator. The result of evaluation is shown respectively. In the sensory evaluation, the transparency of the 16 face images was evaluated by the expert evaluator in the following five stages, and the average value of the evaluation was calculated as the evaluation result for each face image.
1: There is a sense of transparency, 2: There is a little sense of transparency, 3: It can be said neither, 4: There is no sense of transparency, 5: There is no sense of transparency

  According to FIG. 18, the sensory evaluation result of the face image plotted at a position close to the origin is 1, and the numerical value indicating the sensory evaluation result of the face image becomes larger as the plot position is further away from the origin. Yes. According to Spearman's rank correlation coefficient, a correlation value of 0.98 was obtained between the rank from the origin and the rank of the sensory evaluation result. Therefore, in the second embodiment described above, the distance from the origin calculated as indicating the texture of the face appearing in the analysis target face image with respect to the face appearing in the reference face image is the sensory evaluation result. It was verified to show a high correlation. This also proves the validity of quantifying the texture of the entire face in each of the above-described embodiments.

  FIG. 19 is a diagram showing the distance from the origin calculated for each face image shown in FIG. 17 and the sensory evaluation results for each face image. Also in FIG. 19, it can be confirmed that the distance from the origin calculated by the method of the second embodiment described above correlates with the result of the sensory evaluation of the professional evaluator.

  FIG. 20 is a diagram illustrating an example of information indicating a texture analysis result for each range of distance from the origin. From the results as described above, the analysis information generation unit 22 holds information indicating the texture analysis result for each range of distance from the origin as shown in FIG. 20 in advance in the memory 3 and uses this information for analysis. What is necessary is just to produce | generate the texture analysis information of the face reflected in a target face image.

  In addition, each above-mentioned embodiment and each modification can be combined in the range in which the content does not conflict.

  A part or all of each of the above embodiments and modifications may be specified as follows. However, each above-mentioned embodiment and each modification are not restrict | limited to the following description.

<1> sample acquisition means for acquiring average face image information of a plurality of sample face images whose shapes are standardized, and a plurality of basis functions obtained by principal component analysis on the plurality of sample face images;
A principal component analysis means for calculating a basis coefficient of each basis function for the analysis target face image by performing principal component analysis on the analysis target face image whose shape is normalized;
Based on each basis coefficient calculated by the principal component analysis means and a plurality of basis functions and average face image information acquired by the sample acquisition means, each reproduction face corresponding to each of the plurality of cumulative contribution ratios of a predetermined accuracy Reconstruction means for generating each image;
By calculating the difference in light and dark information between the lower-order reproduced face image and the higher-order reproduced face image generated by the reconstructing means with respect to adjacent cumulative contribution rate pairs, the lower-order reproduced face image Difference calculation means for generating at least one of a bright difference image from which dark components have been removed and a dark difference image from which components brighter than the lower-order reproduced face image have been removed for each pair of adjacent cumulative contribution rates When,
At least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component from each color information in the light difference image and the dark difference image is set to each pair of the adjacent cumulative contribution ratios. Representative value calculating means for calculating
By accumulating at least one of a light component representative value and a dark component representative value for each pair of adjacent cumulative contribution rates in ascending order of cumulative contribution rates, a light component cumulative value for each pair of adjacent cumulative contribution rates and Cumulative value calculating means for calculating at least one of dark component cumulative values,
A face image analyzing apparatus.

<2> The cumulative value calculation means obtains the light component cumulative value and the dark component cumulative value in a range of the cumulative contribution ratio from a predetermined lower limit value greater than 50% to a predetermined upper limit value less than 100%.
The face image analyzing apparatus according to <1>.
<3> The predetermined lower limit is 68% or more, the predetermined upper limit is 98% or less, and the predetermined accuracy of the cumulative contribution rate is 3%.
<2> The facial image analysis apparatus according to <2>.
<4> Brightness for each pair of adjacent cumulative contribution ratios calculated by the cumulative value calculation means at coordinates including a first axis indicating the cumulative contribution ratio and a second axis indicating the bright component cumulative value. The adjacent component calculated by the cumulative value calculation means on the light component graph in which the component cumulative value is plotted, and the coordinates including the first axis indicating the cumulative contribution rate and the second axis indicating the dark component cumulative value. First information generating means for generating data for forming at least one of a dark component graph in which dark component cumulative values for each pair of cumulative contributions to be plotted are plotted;
The face image analyzing apparatus according to any one of <1> to <3>, further comprising:
<5> Reference acquisition for acquiring at least one group of a light component cumulative value and a dark component cumulative value for each pair of the adjacent cumulative contribution ratios corresponding to a reference face image to be compared with the analysis target face image means,
Further comprising
The first information generating unit includes the bright component graph in which the bright component cumulative value for each pair of the adjacent cumulative contribution ratios acquired by the reference acquiring unit is superimposed and the reference acquiring unit Generating data to form at least one of the dark component graphs in which dark component cumulative values for each pair of acquired cumulative contributions that are acquired are superimposed and plotted;
<4> The face image analyzing apparatus according to <4>.
<6> Reference acquisition for acquiring at least one group of a light component cumulative value and a dark component cumulative value for each pair of adjacent cumulative contribution ratios corresponding to a reference face image to be compared with the analysis target face image Means,
For each pair of adjacent cumulative contribution ratios, the difference between the bright component accumulated value acquired by the reference acquiring unit and the bright component accumulated value calculated by the accumulated value calculating unit, and acquired by the reference acquiring unit. Comparing means for calculating at least one of the differences between the dark component accumulated value and the dark component accumulated value calculated by the accumulated value calculating means,
The face image analyzer according to any one of <1> to <5>, further comprising:
<7> For each pair of adjacent cumulative contribution ratios calculated by the comparison means at coordinates including a first axis indicating the cumulative contribution ratio and a second axis indicating the difference calculated by the comparison means A light component difference graph in which the difference between the light component accumulated values is plotted, and a dark component difference in which the difference between the dark component accumulated values for each pair of the adjacent cumulative contribution ratios calculated by the comparison unit is plotted. The coordinates of the adjacent cumulative contribution rate calculated by the comparison means are calculated on the coordinates including the graph and the first axis indicating the difference regarding the bright component cumulative value and the second axis indicating the difference regarding the dark component cumulative value. Second information generating means for generating data for forming at least one of comparison graphs in which differences regarding the light component accumulated value and the dark component accumulated value for each pair are plotted;
The face image analyzer according to <6>, further comprising:
<8> calculating the distance from the origin of the position indicated by the difference regarding at least one of the light component cumulative value and the dark component cumulative value calculated by the comparison unit for any one of the adjacent cumulative contribution ratios; Analysis information generating means for generating texture analysis information of a face shown in the analysis target face image based on the calculated distance based on the face shown in the reference face image;
The face image analyzing apparatus according to <6> or <7>, further comprising:
<9> Image acquisition means for acquiring a face image before makeup with a normalized shape and a face image after makeup with a normalized shape as the analysis target face image;
Further comprising
The analysis information generation means generates makeup effect information regarding the pre-makeup face image and the post-makeup face image as the texture analysis information.
<8> The face image analyzing apparatus according to <8>.
<10> Image acquisition means for acquiring a plurality of makeup face images whose shapes are standardized, showing a change in the makeup face over time, as the analysis target face images;
Further comprising
The analysis information generating means generates the makeup disintegration degree information of the makeup face as the texture analysis information.
<8> or <9> The face image analyzer according to <9>.
<11> At least one computer is
Obtaining average face image information of a plurality of sample face images whose shapes are standardized, and a plurality of basis functions obtained by principal component analysis on the plurality of sample face images;
Standardize the shape of the face image to be analyzed,
By performing principal component analysis on the analysis target face image whose shape is normalized, the basis coefficients of the respective basis functions for the analysis target face image are calculated,
Based on each of the calculated basis coefficients and the acquired plurality of basis functions and average face image information, each reproduced face image corresponding to each of the plurality of cumulative contribution ratios of predetermined accuracy is generated,
By calculating the difference in brightness information between the lower-order reproduced face image and the higher-order reproduced face image generated for adjacent cumulative contribution ratio pairs, darker components than the lower-order reproduced face image are removed. Generated at least one of the bright difference image and the dark difference image from which a brighter component is removed than the low-order reproduction face image is generated for each pair of adjacent cumulative contribution rates,
At least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component from each color information in the light difference image and the dark difference image is set to each pair of the adjacent cumulative contribution ratios. For each
By accumulating at least one of a light component representative value and a dark component representative value for each pair of adjacent cumulative contribution rates in ascending order of cumulative contribution rates, a light component cumulative value for each pair of adjacent cumulative contribution rates and Calculating at least one of the dark component cumulative values,
Face image analysis method including
<12> The calculation of the light component cumulative value and the dark component cumulative value is performed within the range of the cumulative contribution ratio from a predetermined lower limit value greater than 50% to a predetermined upper limit value less than 100%, and the light component cumulative value and the dark component cumulative value. Get component cumulative value,
<11> The face image analysis method according to <11>.
<13> The predetermined lower limit is 68% or more, the predetermined upper limit is 98% or less, and the predetermined accuracy of the cumulative contribution rate is 3%.
<12> The face image analysis method according to <12>.
<14> The at least one computer is
A light component graph in which the light component cumulative values calculated for each pair of the adjacent cumulative contribution rates are plotted on coordinates including a first axis indicating a cumulative contribution rate and a second axis indicating the light component cumulative value; And a dark component in which the dark component cumulative value calculated for each pair of adjacent cumulative contribution rates is plotted on coordinates including a first axis indicating the cumulative contribution rate and a second axis indicating the dark component cumulative value. Generating data to form at least one of the graphs;
The face image analysis method according to any one of <11> to <13>, further including:
<15> The at least one computer is
Obtaining a group of at least one of a light component cumulative value and a dark component cumulative value for each pair of the adjacent cumulative contribution ratios corresponding to a reference face image for comparison with the analysis target face image;
The bright component graph in which the bright component cumulative values for each pair of the acquired cumulative contribution ratios acquired are superimposed and the dark component for each pair of the acquired cumulative contribution ratios that are acquired. Generating data to form at least one of the dark component graphs with cumulative values plotted superimposed thereon;
The facial image analysis method according to <14>, further including:
<16> The at least one computer is
Obtaining a group of at least one of a light component cumulative value and a dark component cumulative value for each pair of the adjacent cumulative contribution ratios corresponding to a reference face image for comparison with the analysis target face image;
For each pair of adjacent cumulative contribution ratios, the difference between the bright component cumulative value acquired for the reference face image and the bright component cumulative value calculated for the analysis target face image, and the reference face image is acquired. Calculating at least one of the difference between the dark component accumulated value and the dark component accumulated value calculated for the analysis target face image,
The face image analysis method according to any one of <11> to <15>, further including:
<17> The at least one computer is
The bright component difference in which the difference between the bright component cumulative values for each pair of the adjacent cumulative contribution rates is plotted on the coordinates including the first axis indicating the cumulative contribution rate and the second axis indicating the calculated difference. A graph, a dark component difference graph in which a difference of dark component cumulative values for each pair of adjacent cumulative contribution ratios is plotted, and a first axis indicating a difference related to the light component cumulative value and the dark component cumulative value Forming at least one of a comparison graph in which a difference relating to a light component cumulative value and a dark component cumulative value for each pair of the adjacent cumulative contribution ratios is plotted at coordinates including a second axis indicating a difference regarding Generate data,
The face image analysis method according to <16>, further including:
<18> The at least one computer is
Calculating the distance from the origin of the position indicated by the difference regarding at least one of the light component cumulative value and the dark component cumulative value calculated for any one of the adjacent cumulative contribution ratios;
Based on the calculated distance, generating a texture analysis information of the face shown in the analysis target face image based on the face shown in the reference face image;
The face image analysis method according to <16> or <17>, further including:
<19> The at least one computer is
Obtaining a pre-makeup face image with a normalized shape and a post-makeup face image with a normalized shape as the analysis target face image;
Generating makeup effect information about the pre-makeup face image and the post-makeup face image as the texture analysis information;
The face image analysis method according to <18>, further including:
<20> The at least one computer is
A plurality of makeup face images whose shape is standardized, showing a change in the makeup face over time, are acquired as the analysis target face images;
Generating makeup disintegration degree information of the makeup face as the texture analysis information;
The facial image analysis method according to <18> or <19>, further comprising:
<21> To at least one computer,
Sample acquisition means for acquiring average face image information of a plurality of sample face images whose shapes are normalized, and a plurality of basis functions obtained by principal component analysis on the plurality of sample face images;
A principal component analysis means for calculating a basis coefficient of each basis function for the analysis target face image by performing principal component analysis on the analysis target face image whose shape is normalized;
Based on each basis coefficient calculated by the principal component analysis means and a plurality of basis functions and average face image information acquired by the sample acquisition means, each reproduction face corresponding to each of the plurality of cumulative contribution ratios of a predetermined accuracy Reconstruction means for generating each image;
By calculating the difference in light and dark information between the lower-order reproduced face image and the higher-order reproduced face image generated by the reconstructing means with respect to adjacent cumulative contribution rate pairs, the lower-order reproduced face image Difference calculation means for generating at least one of a bright difference image from which dark components have been removed and a dark difference image from which components brighter than the lower-order reproduced face image have been removed for each pair of adjacent cumulative contribution rates When,
At least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component from each color information in the light difference image and the dark difference image is set to each pair of the adjacent cumulative contribution ratios. Representative value calculating means for calculating
By accumulating at least one of a light component representative value and a dark component representative value for each pair of adjacent cumulative contribution rates in ascending order of cumulative contribution rates, a light component cumulative value for each pair of adjacent cumulative contribution rates and Cumulative value calculating means for calculating at least one of dark component cumulative values,
A program that realizes

2 CPU
3 Memory 4 Input / output I / F
9 Output unit 10 Face image analyzer (analyzer)
DESCRIPTION OF SYMBOLS 11 Image acquisition part 12 Normalization processing part 13 Principal component analysis part 14 Sample data storage part 15 Reconstruction part 16 Difference calculation part 17 Representative value calculation part 18 Cumulative value calculation part 19 Information generation part 20 Reference data storage part 21 Comparison part 22 Analysis information generator

Claims (9)

Sample acquisition means for acquiring average face image information of a plurality of sample face images whose shapes are normalized, and a plurality of basis functions obtained by principal component analysis on the plurality of sample face images;
A principal component analysis means for calculating a basis coefficient of each basis function for the analysis target face image by performing principal component analysis on the analysis target face image whose shape is normalized;
Based on each basis coefficient calculated by the principal component analysis means and a plurality of basis functions and average face image information acquired by the sample acquisition means, each reproduction face corresponding to each of the plurality of cumulative contribution ratios of a predetermined accuracy Reconstruction means for generating each image;
By calculating the difference in light and dark information between the lower-order reproduced face image and the higher-order reproduced face image generated by the reconstructing means with respect to adjacent cumulative contribution rate pairs, the lower-order reproduced face image Difference calculation means for generating at least one of a bright difference image from which dark components have been removed and a dark difference image from which components brighter than the lower-order reproduced face image have been removed for each pair of adjacent cumulative contribution rates When,
At least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component from each color information in the light difference image and the dark difference image is set to each pair of the adjacent cumulative contribution ratios. Representative value calculating means for calculating
By accumulating at least one of a light component representative value and a dark component representative value for each pair of adjacent cumulative contribution rates in ascending order of cumulative contribution rates, a light component cumulative value for each pair of adjacent cumulative contribution rates and Cumulative value calculating means for calculating at least one of dark component cumulative values,
A face image analyzing apparatus. The cumulative value calculation means obtains the light component cumulative value and the dark component cumulative value within a range of the cumulative contribution ratio from a predetermined lower limit value greater than 50% to a predetermined upper limit value less than 100%.
The face image analysis apparatus according to claim 1. The predetermined lower limit is 68% or more, the predetermined upper limit is 98% or less, and the predetermined accuracy of the cumulative contribution rate is 3%.
The face image analysis apparatus according to claim 2. Reference acquisition means for acquiring a group of light component accumulated values and dark component accumulated values for each pair of the adjacent accumulated contribution rates corresponding to a reference face image for comparison with the analysis target face image;
The light component cumulative value calculated by the cumulative value calculation means for each pair of the adjacent cumulative contribution rates on the coordinates including the first axis indicating the cumulative contribution rate and the second axis indicating the bright component cumulative value, and The light component graph obtained by superimposing the light component cumulative value acquired by the reference acquisition unit, and the coordinates including the first axis indicating the cumulative contribution rate and the second axis indicating the dark component cumulative value, At least one of the dark component graph in which the dark component accumulated value calculated by the accumulated value calculating unit and the dark component accumulated value acquired by the reference acquiring unit for each pair of the adjacent accumulated contribution rates are plotted in a superimposed manner. First information generating means for generating data for forming one;
The face image analyzing apparatus according to claim 3, further comprising: Reference acquisition means for acquiring at least one group of a light component cumulative value and a dark component cumulative value for each pair of the adjacent cumulative contribution ratios corresponding to a reference face image for comparison with the analysis target face image;
For each pair of adjacent cumulative contribution ratios, the difference between the bright component accumulated value acquired by the reference acquiring unit and the bright component accumulated value calculated by the accumulated value calculating unit, and acquired by the reference acquiring unit. Comparing means for calculating at least one of the differences between the dark component accumulated value and the dark component accumulated value calculated by the accumulated value calculating means,
The face image analysis apparatus according to any one of claims 1 to 4, further comprising: A bright component for each pair of the adjacent cumulative contribution ratios calculated by the comparison means at coordinates including a first axis indicating the cumulative contribution ratio and a second axis indicating the difference calculated by the comparison means. A bright component difference graph in which a difference in cumulative values is plotted, a dark component difference graph in which a difference in dark component cumulative values for each pair of adjacent cumulative contribution rates calculated by the comparison unit is plotted, and For each pair of the adjacent cumulative contribution ratios calculated by the comparison means at coordinates including a first axis indicating a difference regarding the light component cumulative value and a second axis indicating a difference regarding the dark component cumulative value. Second information generating means for generating data for forming at least one of comparison graphs in which differences regarding the light component accumulated value and the dark component accumulated value are plotted;
The face image analyzing apparatus according to claim 5, further comprising: The distance from the origin of the position indicated by the difference related to at least one of the light component cumulative value and the dark component cumulative value calculated by the comparison unit for any one of the adjacent cumulative contribution ratios is calculated, and the calculated Analysis information generating means for generating texture analysis information of a face shown in the analysis target face image based on a distance based on a face shown in the reference face image;
The face image analyzing apparatus according to claim 5 or 6, further comprising: At least one computer
Obtaining average face image information of a plurality of sample face images whose shapes are standardized, and a plurality of basis functions obtained by principal component analysis on the plurality of sample face images;
Standardize the shape of the face image to be analyzed,
By performing principal component analysis on the analysis target face image whose shape is normalized, the basis coefficients of the respective basis functions for the analysis target face image are calculated,
Based on each of the calculated basis coefficients and the acquired plurality of basis functions and average face image information, each reproduced face image corresponding to each of the plurality of cumulative contribution ratios of predetermined accuracy is generated,
By calculating the difference in brightness information between the lower-order reproduced face image and the higher-order reproduced face image generated for adjacent cumulative contribution ratio pairs, darker components than the lower-order reproduced face image are removed. Generated at least one of the bright difference image and the dark difference image from which a brighter component is removed than the low-order reproduction face image is generated for each pair of adjacent cumulative contribution rates,
At least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component from each color information in the light difference image and the dark difference image is set to each pair of the adjacent cumulative contribution ratios. For each
By accumulating at least one of a light component representative value and a dark component representative value for each pair of adjacent cumulative contribution rates in ascending order of cumulative contribution rates, a light component cumulative value for each pair of adjacent cumulative contribution rates and Calculating at least one of the dark component cumulative values,
Face image analysis method including On at least one computer,
Sample acquisition means for acquiring average face image information of a plurality of sample face images whose shapes are normalized, and a plurality of basis functions obtained by principal component analysis on the plurality of sample face images;
A principal component analysis means for calculating a basis coefficient of each basis function for the analysis target face image by performing principal component analysis on the analysis target face image whose shape is normalized;
Based on each basis coefficient calculated by the principal component analysis means and a plurality of basis functions and average face image information acquired by the sample acquisition means, each reproduction face corresponding to each of the plurality of cumulative contribution ratios of a predetermined accuracy Reconstruction means for generating each image;
By calculating the difference in light and dark information between the lower-order reproduced face image and the higher-order reproduced face image generated by the reconstructing means with respect to adjacent cumulative contribution rate pairs, the lower-order reproduced face image Difference calculation means for generating at least one of a bright difference image from which dark components have been removed and a dark difference image from which components brighter than the lower-order reproduced face image have been removed for each pair of adjacent cumulative contribution rates When,
At least one of a light component representative value indicating the intensity of the light component and a dark component representative value indicating the intensity of the dark component from each color information in the light difference image and the dark difference image is set to each pair of the adjacent cumulative contribution ratios. Representative value calculating means for calculating
By accumulating at least one of a light component representative value and a dark component representative value for each pair of adjacent cumulative contribution rates in ascending order of cumulative contribution rates, a light component cumulative value for each pair of adjacent cumulative contribution rates and Cumulative value calculating means for calculating at least one of dark component cumulative values,
A program that realizes