JP5818714B2 - Makeup face image evaluation apparatus and makeup face image evaluation method - Google Patents

Description

  The present invention relates to a technique for quantitative evaluation of a face (makeup face) image on which makeup has been applied.

  There has been proposed a method for objectively analyzing the state and appearance of skin using image analysis technology.

  In Patent Document 1 below, physical glossiness and apparent roughness of the skin surface are acquired based on image data obtained by multi-resolution analysis using wavelet transform, and skin gloss is analyzed from these information. A method has been proposed. Patent Documents 2 and 3 below propose a method of synthesizing face images having different stereoscopic effects and appearance ages without changing the basic design of the face such as the contour of the face using principal component analysis. Non-Patent Document 1 below describes sensory evaluation of appearance age using five standard faces created for each age. In addition, this standard face statistically processes a large number of real images to extract average spots and wrinkles, add them to the average face, and further change the roughness of the texture equivalent to the year. It is described that it will be completed.

JP 2004-166801 A JP 2008-276405 A JP 2009-294958 A

“Mandom, Study on“ Appearance Impression ”-Creation and Application of“ Standard Face ””, Mandom Co., Ltd., August 18, 2009

  However, at present, there is no objective evaluation method for makeup, and subjective evaluation must be relied upon when evaluating the quality of makeup, makeup status, and the like. On the other hand, various types of cosmetics and makeup methods are provided. Therefore, a method for objectively evaluating the makeup effect is strongly desired in order for each individual to grasp whether the cosmetics are being used correctly or whether the makeup method is correctly performed.

  The present invention has been made in view of such problems, and provides a technique for quantitatively evaluating a makeup face based on a makeup face image.

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

  The first aspect relates to a makeup face image evaluation apparatus. The makeup face image evaluation apparatus according to the first aspect performs independent component analysis on a plurality of sets of sample face images and sample makeup face images each having a standardized shape, whereby each sample face image and each sample makeup face image are analyzed. A plurality of basis vectors based on independent component analysis means for calculating coefficients of a plurality of basis vectors, and a degree of variation in coefficients between sample face images and a degree of variation in coefficients between sample makeup face images in each basis vector. A selection means for selecting a part of the image as a plurality of evaluation basis vectors, an evaluation image vector obtained from the makeup face image to be evaluated, a target image vector obtained from the target makeup face image, and the selection means. Evaluation information for obtaining evaluation information of the makeup face image to be evaluated from the inner product with each evaluation basis vector Out and means.

  The second aspect relates to a makeup face image evaluation method. The makeup face image evaluation method according to the second aspect performs independent component analysis on a plurality of sets of sample face images and sample makeup face images each having a standardized shape, whereby each sample face image and each sample makeup face image are analyzed. Calculating a coefficient of each of the plurality of basis vectors, and based on the coefficient variation degree between the sample face images and the coefficient variation degree between the sample makeup face images in each basis vector, a part of the plurality of basis vectors Are selected as a plurality of evaluation basis vectors, and the evaluation image vector obtained from the makeup face image to be evaluated and the inner product of the target image vector obtained from the target makeup face image and each evaluation basis vector are used to determine the evaluation target vector. Including obtaining evaluation information of a makeup face image.

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

  According to each aspect described above, it is possible to provide a technique for quantitatively evaluating a makeup face based on a makeup face image.

It is a figure which shows the intrinsic | native component (base) of the face extracted by PCA. It is a figure which shows the independent component (base) of the face extracted by ICA. It is a figure which shows the change of the face image based on operation of ICA base. It is a figure which shows the comparative example of a natural face image and a makeup | decoration face image. It is a figure which shows notionally the hardware structural example of the makeup | decoration face image evaluation apparatus in 1st Embodiment. It is a figure which shows notionally the process structural example of the makeup | decoration face image evaluation apparatus in 1st Embodiment. It is a graph which shows the example of the difference of the coefficient dispersion | variation between the sample face images in each base, and the coefficient dispersion | distribution between sample makeup face images. It is a figure which shows the example of the base image corresponding to the basis vector for evaluation. It is a flowchart which shows the operation example of the learning processing system of the makeup | decoration face image evaluation apparatus in 1st Embodiment. It is a flowchart which shows the operation example of the evaluation processing system of the makeup | decoration face image evaluation apparatus in 1st Embodiment. It is a figure which shows the evaluation object makeup | decoration face image for 10 persons used in each Example, and its natural-face image. It is a figure which shows the target makeup | decoration face image used in each Example. It is a figure which shows the evaluation information (function value) obtained in each Example. It is a figure which shows the result of a questionnaire survey. It is a figure which shows the comparison result of the questionnaire survey result and the evaluation information (function value) obtained in each Example.

  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 makeup face image evaluation apparatus according to the present embodiment performs independent component analysis on a plurality of sets of sample face images and sample makeup face images each having a standardized shape, whereby each sample face image and each sample makeup face image are analyzed. An independent component analysis unit that calculates coefficients of a plurality of basis vectors, and a plurality of basis vectors based on a degree of coefficient variation between sample face images and a coefficient variation degree between sample makeup face images in each basis vector. Are selected by a selection unit that selects a part of the image as a plurality of evaluation basis vectors, an evaluation image vector obtained from the makeup face image to be evaluated, and a target image vector obtained from the target makeup face image and the selection unit. Evaluation information calculation for obtaining evaluation information of the makeup face image to be evaluated from the inner product with each evaluation basis vector Provided with a door.

  Here, the bare face means an unmake-up face in a state where at least a part of the cosmetic used for the makeup face is not applied, and is limited to a bare skin state where no cosmetic is applied. Is not to be done. The makeup face means a face to which some kind of cosmetic is applied. The target makeup face means a makeup face that is a target to which some cosmetic is applied in order to achieve a certain purpose. For example, the target makeup face is an ideal makeup face that has been made up by a professional makeup artist, a makeup face that has been made up to achieve a certain impression (gorgeous impression, feminine impression, etc.) In addition, it is a makeup face or the like with a desired makeup to be realized. The target makeup face may be anything as long as it can be compared with the makeup face to be evaluated. The face image may be an image of a user's face or an image of another person's face viewed from the user including a celebrity.

  In addition, the sample face image and the sample makeup face image whose shape is standardized are subjected to normalization processing such as normalization on each sample image, respectively, so that the shape characteristics of the face of each sample provider (hair ), A facial image in which the contours of the cheek or chin, eyebrows, eyes, nose, etc. are removed. In other words, each sample face image and each sample makeup face image whose shape is standardized include a predetermined average face shape and a texture unique to each sample provider. For the normalization process, a warping process, a morphing process, or the like using Piece-wise Affine is used. Since such a standardization process may be performed by using a well-known technique, a description thereof is omitted here.

In the makeup face image evaluation apparatus, coefficients of a plurality of base vectors are calculated for each of the sample face image and the sample makeup face image whose shapes are standardized by independent component analysis (ICA). The Independent component analysis (ICA) is a method of decomposing a linear sum of a plurality of basis vectors (basis functions) common to a plurality of sample images (plain face image and makeup face image) whose shapes are standardized. Since the algorithm of independent component analysis is described in Reference Document 1 below, description thereof is omitted here.
Reference 1: Bell, AJ and Sejnowski, TJ, "An information-maximization approach to blind separation and blind deconvolution", Neural Computation 7, 1129-1159 (1995).

  In such an independent component analysis, a face image whose shape is standardized is used, and thus the shape features of the face such as the face contour are discarded from each obtained basis vector. . Then, according to the independent component analysis, local texture information forming the face is extracted as each basis vector. The characteristics of such independent component analysis (ICA) will be described below in comparison with principal component analysis (PCA).

  PCA finds an intrinsic component (base) having a large variance, and ICA finds an independent component (base). Further, the basis obtained by PCA has a spatial globality, and the basis obtained by ICA has a spatial locality and directionality.

  FIG. 1A shows an eigen component (base) of a face extracted by PCA, and FIG. 1B shows an independent component (base) of a face extracted by ICA. Hereinafter, a component extracted by ICA may be expressed as an ICA component or an ICA base, and a component extracted by PCA may be expressed as a PCA component or a PCA base. According to FIG. 1A and FIG. 1B, each ICA base is independent from each other, and thus is viewed in a different form. Furthermore, each ICA base is responsive to local parts of the face such as the nose, cheeks and forehead.

  As described above, when the face image is analyzed using ICA, it is possible to visually recognize what the obtained components are reacting to and what is conspicuous. That is, each ICA base represents each texture of the face grasped by human vision and is considered to have texture information of the local part of the face. Based on this idea, it is considered that the facial texture can be changed by manipulating a specific ICA basis.

  FIG. 2 is a diagram illustrating changes in the face image based on an operation based on ICA. P1 indicates a face image of a person with a stain under the right eye. P2 indicates one ICA base (30th) among 88 ICA bases. This ICA base corresponds to a component representing a stain under the right eye. P3 indicates a face image obtained by reconstructing all ICA bases without manipulating the ICA bases. P4 indicates a face image obtained by reconstructing by removing the ICA base of P2. According to P4, compared to the original face image P1, it can be seen that the stain under the right eye has disappeared.

From this, each component (base) extracted by ICA has facial texture information, and the facial texture or makeup effect can be quantified by using the size of each component (coefficient of each base vector). Is proved. Note that the reconstruction of the ICA base in P3 and P4 in FIG. 2 is realized by an algorithm described in Reference Document 2 below, for example.
Reference 2: Chen Yan Wei, “Application of Independent Component Analysis (ICA) to Pattern Recognition and Image Processing and MATLAB Lab Simulation”, Trikes, Inc., October 31, 2007

  In the present embodiment, the basis vectors are used for evaluation from a plurality of basis vectors based on the coefficient variation degree between the sample face images and the coefficient variation degree between the sample makeup face images. A plurality of evaluation basis vectors are selected.

  FIG. 3 is a diagram illustrating a comparative example of a bare face image and a makeup face image. According to FIG. 3, the makeup face image suppresses spots, wrinkles, shine, and the like by makeup as compared with the face image, and the skin is visually recognized as a whole. That is, in the case of a natural face, the texture differs greatly depending on the person, so the coefficient indicating the size of each independent component (base) also differs greatly for each face image. On the other hand, in the case of a makeup face, since the skin becomes uniform as a whole and the difference in texture becomes small, the difference in coefficient of each independent component (base) by a person also becomes small. In other words, the degree of variation in the coefficient of each base vector is large between the face images, and the degree of variation in the coefficient of each base vector is small between the makeup face images. Therefore, a base vector having a large difference between the coefficient variation degree between the face images and the coefficient variation degree between the makeup face images can be considered as a component having a large makeup effect.

  In the present embodiment, as described above, a plurality of evaluation basis vectors used for evaluation are selected based on the degree of coefficient variation. Thereby, according to the present embodiment, a basis vector representing a texture having a large makeup effect can be selected as an evaluation basis vector.

  In the present embodiment, the evaluation is performed based on the inner product of the evaluation image vector obtained from the makeup face image to be evaluated, the target image vector obtained from the target makeup face image, and each evaluation base vector selected by the selection unit. Evaluation information of the target makeup face image is acquired. According to the inner product of the evaluation image vector and each evaluation base vector, the coefficient of each evaluation base vector that the evaluation image vector has is calculated, and according to the inner product of the target image vector and each evaluation base vector, the target image vector is The coefficient of each evaluation basis vector is calculated.

  As described above, since each evaluation basis vector represents a texture having a large makeup effect, for example, if each of the above-described coefficients is used as the evaluation information, a makeup face image and a target to be evaluated for the texture representing the makeup effect, The index value (feature value) of the makeup face image can be obtained. For example, when both coefficients are approximated with respect to each evaluation basis vector, it can be considered that the makeup face image to be evaluated is close to the target makeup face image. The fact that it is close to the target makeup face image can be considered that the makeup face image to be evaluated is makeup according to the target.

  Furthermore, the distance of each coefficient obtained from the inner product may be calculated. Thus, according to this embodiment, the index value about each makeup face image can be obtained, and the makeup face can be quantitatively evaluated based on the makeup face image.

  Hereinafter, the details of the above-described embodiment will be described.

〔Device configuration〕
FIG. 4 is a diagram conceptually illustrating a hardware configuration example of the makeup face image evaluation apparatus 1 in the first embodiment. The makeup face image evaluation apparatus 1 according to the first embodiment is a so-called computer. 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. Have 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 user interface device such as a display device or an input device. The input / output I / F 4 may be connected to a communication device or the like that communicates with another computer via a network (not shown). The hardware configuration of the makeup face image evaluation apparatus 1 is not limited.

  FIG. 5 is a diagram conceptually illustrating a processing configuration example of the makeup face image evaluation apparatus 1 in the first embodiment. The makeup face image evaluation apparatus 1 according to the first embodiment includes a learning processing system 10 and an evaluation processing system 20. The learning processing system 10 specifies the plurality of evaluation basis vectors representing the makeup effect by learning using a plurality of sets of sample face images and sample makeup face images. The evaluation processing system 20 uses the plurality of evaluation basis vectors obtained by the learning processing system 10 to calculate evaluation information of the makeup face image to be evaluated with reference to the target makeup face image.

  The learning processing system 10 includes a sample image acquisition unit 11, a normalization unit 13, a vectorization unit 14, an independent component analysis unit 15, a variance calculation unit 17, and a selection unit 19. The evaluation processing system 20 includes a target image acquisition unit 21, a normalization unit 22, a vectorization unit 23, an evaluation information calculation unit 25, and an evaluation information output unit 27. 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.

<< Learning processing system 10 >>
The sample image acquisition unit 11 acquires image data of a plurality of sets of sample face images and sample makeup face images. Each image data is acquired from a portable recording medium, another computer or the like via the input / output I / F 4. The plural sets of sample face images and sample makeup face images are the face images and makeup face images of each sample provider. Hereinafter, the sample face image and the sample makeup face image may be simply referred to as a sample image except when it is necessary to distinguish them.

  Although it is desirable to use an image of bare skin to which no cosmetic is applied, the face image to which only the base cosmetic is applied may be used as the bare face image. For the makeup face image, each sample provider's favorite makeup method and the image of each sample provider's face to which cosmetics have been applied may be used, or each of the sample makeup methods and cosmetics to which a predetermined makeup method has been applied. An image of the sample provider's face may be used. Further, the number of sets of the sample face image and the sample makeup face image is not limited, but the evaluation accuracy increases as the number increases.

  The normalization unit 13 performs normalization processing such as normalization on each sample image acquired by the sample image acquisition unit 11, thereby obtaining each sample face image and each sample makeup face image whose shapes are normalized. get. The normalization process is as described above.

  The vectorization unit 14 vectorizes each sample image whose shape has been normalized by the normalization unit 13. The vectorization unit 14 generates a sample image vector for each sample image by this vectorization. For example, raster scanning is used for vectorization.

  The independent component analysis unit 15 performs independent component analysis on each sample image vector generated by the vectorization unit 14. The independent component analysis unit 15 decomposes each sample image into linear sums of a plurality of basis vectors common to the sample images by this independent component analysis. As a result, the independent component analysis unit 15 calculates a plurality of basis vectors common to the sample images and coefficients of the plurality of basis vectors related to each sample face image and each sample makeup face image.

  The variance calculation unit 17 calculates the variance of the coefficients between the sample face images and the variance of the coefficients between the sample makeup face images for each basis vector calculated by the independent component analysis unit 15. Thus, in the first embodiment, variance is used as an index indicating the degree of coefficient variation.

  Based on the difference between the coefficient variance between the sample face images and the coefficient variance between the sample makeup face images in each basis vector, the selection unit 19 includes a plurality of basis vectors calculated by the independent component analysis unit 15. A part is selected as a plurality of evaluation basis vectors.

  FIG. 6 is a graph showing an example of the difference between the coefficient variance between the sample face images and the coefficient variance between the sample makeup face images in each base. In FIG. 6, the horizontal axis represents each base, and the vertical axis represents the variance difference. For example, the selection unit 19 selects 12 basis vectors from the 80 basis vectors shown in FIG. 6 in descending order of the variance difference. In the example of FIG. 6, there is a base in which the variance difference has a negative value. However, the fact that the variance of the makeup face image is larger than the variance of the bare face image is a phenomenon that hardly occurs in actual makeup.

  FIG. 7 is a diagram illustrating an example of a base image corresponding to an evaluation base vector. FIG. 7 shows twelve base images selected in descending order of the variance difference based on the variance difference of FIG. The number given on each base image indicates the number of the base vector. The base image at the left end is the base image having the largest variance difference and indicates a component having a large makeup effect.

  In the example of FIG. 7, a case where 12 evaluation basis vectors are selected is illustrated, but in this embodiment, the number of evaluation basis vectors to be selected is not limited. The selection unit 19 may hold a predetermined number as the number of evaluation basis vectors, hold a predetermined threshold value, and select a basis vector having a variance difference larger than the predetermined threshold value as an evaluation basis vector. . However, if the number of evaluation basis vectors is too small, the accuracy of the obtained evaluation information is reduced, and if all the basis vectors are used as the evaluation basis vectors, the calculation speed of the evaluation information may be reduced. There is. Therefore, the number of evaluation basis vectors is preferably set to a number that maintains the accuracy of the evaluation information and obtains a practical calculation speed. For example, it is more preferable that 5 or more and 20 or less evaluation basis vectors are used.

<< Evaluation Processing System 20 >>
The target image acquisition unit 21 acquires image data of the makeup face image to be evaluated and the target makeup face image. Each image data may be acquired from a portable recording medium, another computer, or the like via the input / output I / F 4 or acquired from an imaging unit (not shown) connected to the input / output I / F 4. May be.

  The normalization unit 22 performs a normalization process such as normalization on each image data acquired by the target image acquisition unit 21, so that the makeup face image to be evaluated whose shape is standardized and the shape is standardized The target makeup face image is obtained. The normalization process is as described above. Note that the target image acquisition unit 21 may acquire each image data whose shape has already been standardized. By using standardized face images in this way, each face image of the makeup face image to be evaluated and the target makeup face image can be of the same person or different people, and in both cases comparative evaluation It becomes possible to do.

  The vectorization unit 23 vectorizes each image data whose shape has been normalized by the normalization unit 22, thereby obtaining an evaluation image vector corresponding to the makeup face image to be evaluated and a target image corresponding to the target makeup face image. Generate a vector. For example, raster scanning is used for vectorization.

The evaluation information calculation unit 25 applies the evaluation base vector, the evaluation image vector, and the target image vector selected by the selection unit 19 to the evaluation function D (x) (see (Equation 1) below), thereby The distance between the coefficient of the evaluation image vector and the coefficient of the target image vector in the base vector is calculated as the evaluation information of the evaluation target makeup face image. In the following (Equation 1), x represents an evaluation image vector, x ₀ represents a target image vector, u _i represents an _i- th evaluation basis vector, k represents the number of evaluation basis vectors, and T Indicates transposition.

In the above (Formula 1), by an inner product between u _i and x _0, the coefficient is obtained about the target image vector in the evaluation basis vector u _i, the inner product of the u _i and x, evaluation in the evaluation basis vector u _i A coefficient for the image vector is obtained. That is, the above (Expression 1) is an expression representing the distance between the coefficient of the evaluation image vector and the coefficient of the target image vector in each evaluation base vector. In other words, according to (Equation 1) above, between the projections in the case where the evaluation image vector and the target image vector are projected onto the subspace representing the makeup effect composed of the evaluation basis vectors selected by the selection unit 19. The Euclidean distance is calculated.

  The evaluation information output unit 27 determines the proximity of the evaluation target makeup face image to the target makeup face image based on the distance (evaluation information) calculated by the evaluation information calculation unit 25, and the determination result and the evaluation information Result information including at least one is output in a predetermined form. The distance D calculated by the evaluation information calculation unit 25 represents the closeness of the evaluation target makeup face image to the target makeup face image in the texture component having a large makeup effect. That is, it can be determined that the makeup is made so as to approach the target makeup face as the distance D is small, and conversely, the makeup is applied so that the makeup is different from the target makeup face as the distance D is large. Can be determined.

  Specifically, the evaluation information output unit 27 holds in advance a correspondence table between the distance range and determination data indicating the proximity of the evaluation target makeup face image to the target makeup face image. Determination data corresponding to the calculated distance is extracted from the correspondence table, and the extracted determination data is included in the result information. In this case, for example, the distance range (less than 2.0) and the determination data (A; good) are associated with the correspondence table, and the distance range (2.0 to 2.5) and the determination data (B; normal). Are associated with each other, the distance range (2.0 to 3.0) is associated with the determination data (B; normal), and the distance range (3.0 or more) is associated with the determination data (C; poor).

  The output form may be a form in which a file including the information as text data is generated, a form in which display data for displaying the information on a display device is generated, It may be in a form in which print data for causing the printing apparatus to print information is generated. The present embodiment does not limit the output form of information indicating such a determination result.

[Operation example]
Hereinafter, the makeup face image evaluation device 1 is simply referred to as the evaluation device 1, and an operation example of the evaluation device 1 will be described with reference to FIGS. FIG. 8 is a flowchart illustrating an operation example of the learning processing system 10 of the makeup face image evaluation apparatus 1 according to the first embodiment.

  The evaluation apparatus 1 acquires image data of a plurality of sets of sample face images and sample makeup face images (S81).

  Subsequently, the evaluation apparatus 1 performs a normalization process on each sample image (S82). As a result, the evaluation apparatus 1 acquires a standardized image (each sample face image and each sample makeup face image whose shape is standardized) in which the shape characteristics of the face are discarded.

  The evaluation apparatus 1 vectorizes each sample image whose shape is standardized (S83). Thereby, the evaluation apparatus 1 generates each sample image vector corresponding to each sample face image and each sample makeup face image.

  Subsequently, the evaluation apparatus 1 performs independent component analysis on each sample image vector (S84). Thereby, the evaluation apparatus 1 calculates the coefficients of the plurality of basis vectors for each sample image. Each base vector common to each sample image represents an independent face texture.

  The evaluation device 1 calculates the coefficient variance between the sample face images and the coefficient variance between the sample makeup face images for each basis vector calculated by the independent component analysis (S85).

  The evaluation apparatus 1 uses a part of a plurality of basis vectors calculated by independent component analysis based on the difference between the coefficient variance between the sample face images and the coefficient variance between the sample makeup face images in each basis vector. Are selected as a plurality of evaluation basis vectors (S86).

  FIG. 9 is a flowchart showing an operation example of the evaluation processing system 20 of the makeup face image evaluation apparatus 1 in the first embodiment.

  The evaluation apparatus 1 acquires each image data of the makeup face image to be evaluated and the target makeup face image (S91).

  The evaluation apparatus 1 performs a normalization process on each acquired image data (S92). Thereby, the evaluation apparatus 1 acquires the makeup face image to be evaluated and the target makeup face image whose shape is standardized.

  Subsequently, the evaluation device 1 vectorizes the evaluation target makeup face image and the target makeup face image whose shape is standardized (S93). Thereby, the evaluation apparatus 1 generates an evaluation image vector corresponding to the makeup face image to be evaluated and a target image vector corresponding to the target makeup face image.

  The evaluation apparatus 1 applies the evaluation basis vector (selected in the process (S86) in the example of FIG. 8), the evaluation image vector, and the target image vector to the evaluation function D (x) (see (Expression 1) above). Then, evaluation information of the evaluation target makeup face image is calculated (S94). The distance between the coefficient of the evaluation image vector and the coefficient of the target image vector in each evaluation base vector is calculated by the evaluation function D (x).

  The evaluation device 1 determines the proximity of the evaluation target makeup face image to the target makeup face image using the calculated evaluation information (S95). The determination result may be a symbol (A, B, etc.) that can specify the determination content, or may be a text that indicates the determination content.

  The evaluation apparatus 1 outputs result information including at least one of the determination result of (S95) and the evaluation information of (S94) in a predetermined form (S96).

  Such an operation of the evaluation processing system 20 may be started after the selection of the evaluation basis vector in the learning processing system 10 (S86 in FIG. 8), or is performed independently of the operation of the learning processing system 10. May be. Further, since the evaluation basis vector is used in the processing (S94), the learning base system 10 selects the evaluation basis vector (S86 in FIG. 8) before the processing (S94) is executed. The operations of the processing system 10 and the evaluation processing system 20 may be controlled. Further, the operation of the learning processing system 10 shown in FIG. 8 may be executed each time a set of a sample face image and a sample makeup face image is acquired, or after selecting an evaluation basis vector (S86), When a new at least one set of sample face image and sample makeup face image has been acquired, it may be executed at an arbitrary timing.

[Operation and Effect of First Embodiment]
As described above, in the first embodiment, coefficients of a plurality of base vectors are calculated for each of the sample bare face image and the sample makeup face image whose shapes are standardized by independent component analysis (ICA). Then, the coefficient variance between the sample face images and the coefficient variance between the sample makeup face images are calculated for each basis vector, and a basis vector having a large coefficient variance difference is selected as an evaluation basis vector.

  The plurality of basis vectors acquired in the first embodiment each represent local texture information forming a face, and each evaluation basis vector selected from them has a large coefficient variance difference. Among local textures, it represents a texture with a large makeup effect. As described above, according to the first embodiment, by using independent component analysis (ICA), it is possible to extract a component representing a facial texture and specify a component having a large makeup effect. In calculating the evaluation information of the evaluation target makeup face image, only components having a large makeup effect can be used.

  That is, in the first embodiment, the coefficient of each evaluation base vector included in the evaluation image vector corresponding to the evaluation target makeup face image and the coefficient of each evaluation base vector included in the target image vector corresponding to the target makeup face image. Is calculated by the evaluation function D (x) as evaluation information of the evaluation target makeup face image. Finally, the proximity of the evaluation target makeup face image to the target makeup face image is determined using this distance, and result information including at least one of the determination result and the distance is output.

  As described above, according to the first embodiment, the difference between the feature amounts of the evaluation target makeup face image and the target makeup face image in the texture having a large makeup effect can be calculated as the inter-coefficient distance. Thus, the quantitative evaluation of the makeup face such as the closeness of the evaluation target makeup face image to the target makeup face image can be realized.

[Modification]
The makeup face image evaluation apparatus 1 in the first embodiment described above does not have the sample image acquisition unit 11, the normalization unit 13, the vectorization unit 14, and the independent component analysis unit 15, but via a network or a portable recording medium. Alternatively, a plurality of base vectors common to the sample image and coefficients of the plurality of base vectors related to each sample bare face image and each sample makeup face image may be acquired from the outside. In this case, the operation of the learning processing system 10 shown in FIG. 8 is to execute a process of acquiring a plurality of basis vectors and the coefficients instead of (S81), (S82), (S83), and (S84). Good.

  Similarly, the makeup face image evaluation apparatus 1 does not have the target image acquisition unit 21, the normalization unit 22, and the vectorization unit 23, and converts the makeup face image evaluation device from the outside to the evaluation target makeup face image via a network or a portable recording medium. A target image vector corresponding to the corresponding evaluation image vector and the target makeup face image may be acquired. In this case, the operation of the evaluation processing system 20 shown in FIG. 9 may be executed by acquiring the evaluation image vector and the target image vector instead of (S91), (S92), and (S93).

  Examples will be given below to describe the above-described embodiment in more detail. The present invention is not limited in any way by the following examples. According to each of the following examples, the effectiveness of the evaluation information (value obtained by the evaluation function D (x)) obtained in the above-described embodiment was verified by comparing with the result of the questionnaire survey.

  FIG. 10 is a diagram illustrating the evaluation subject makeup face images and the bare face images for ten people used in each example. FIG. 11 is a diagram illustrating a target makeup face image used in each example. The images shown in FIGS. 10 and 11 were obtained by photographing a face illuminated with a halogen light from each direction of +15 degrees and −15 degrees from the front of the face with a digital camera (Nikon D70) from the front of the face. It is an image. In addition, as the target makeup face image of FIG. 11, a makeup face image on which makeup is applied by a professional makeup artist is used. The evaluation target makeup face image and the target makeup face image are not limited to those acquired by the above-described photographing method. For example, an image randomly extracted from a database storing a plurality of multi-viewpoint / multi-illumination makeup face images may be used as the evaluation target makeup face image.

  Example 1 shows an example in which eight (k = 8) evaluation basis vectors are selected. Example 2 shows an example in which ten (k = 10) evaluation basis vectors are selected. Example 3 shows an example in which 12 (k = 12) evaluation basis vectors are selected. In each example, 41 face images and makeup face images (82 images in total) were used as sample face images and sample makeup face images.

  FIG. 12 is a diagram showing evaluation information (function values) obtained in each example. According to FIG. 12, although the 2nd place, the 3rd place, and the 4th place differ in each example, it turns out that there is no big difference in the evaluation result in each example.

On the other hand, the questionnaire survey was conducted as follows.
(Survey object) 15 men and women in their 20s to 30s (10 men and 5 women)
(Investigation method) As shown in FIG. 10, ten sheets of paper (A to J) printed in a state in which the face image and the makeup face image are arranged are shown to each investigator, and according to the ranking of evaluation. Sort the paper.
(Evaluation Criteria) A person who feels that skin, wrinkles, and shine have been reduced before and after makeup and that the skin is clean is ranked high.

  FIG. 13 is a diagram showing the results of a questionnaire survey. In the table of FIG. 13, the numbers of surveyors selected for each rank are shown for A to J. The final ranking obtained by the questionnaire survey is shown in the bottom line of FIG. As for the seventh place, E and F have the same number (4), but since F is fixed at the eighth place, E is determined as the seventh place.

  FIG. 14 is a diagram illustrating a comparison result between the questionnaire survey result and the evaluation information (function value) obtained in each example. According to FIG. 14, although there is a slight difference, it is proved that there is a correlation between the evaluation information obtained in each example and the questionnaire survey result. In particular, the evaluation information obtained in Example 1 (k = 8) and the questionnaire survey result show a high correlation.

  From such a comparison result, it is proved that the evaluation information obtained in each embodiment, that is, the value obtained by the evaluation function D (x) has a correlation with the evaluation based on human sensitivity. According to the embodiment, it was proved that a makeup face image can be quantitatively evaluated.

1 Makeup face image evaluation device (evaluation device)
2 CPU
3 Memory 4 Input / output I / F
DESCRIPTION OF SYMBOLS 10 Learning processing system 11 Sample image acquisition part 13, 22 Normalization part 14, 23 Vectorization part 15 Independent component analysis part 17 Variance calculation part 19 Selection part 20 Evaluation processing system 21 Target image acquisition part 25 Evaluation information calculation part 27 Evaluation information Output section

Claims (7)

Independent calculation of multiple base vector coefficients for each sample face image and each sample face image by independent component analysis of multiple sets of sample face images and sample face images that have been normalized in shape Component analysis means;
Based on the coefficient variation degree between the sample face images and the coefficient variation degree between the sample makeup face images in each of the basis vectors, a part of the plurality of basis vectors is used as a plurality of evaluation basis vectors. A selection means to select;
From the inner product of the evaluation image vector obtained from the makeup face image to be evaluated, the target image vector obtained from the target makeup face image, and the respective basis vectors for evaluation selected by the selection means, the makeup face image to be evaluated Evaluation information calculation means for acquiring evaluation information of
A makeup face image evaluation apparatus comprising: The evaluation information calculating means calculates a distance between coefficients of the evaluation image vector and the target image vector obtained by the inner product as the evaluation information;
The makeup face image evaluation apparatus according to claim 1. For each of the basis vectors, further comprises a variance calculating means for calculating the variance of the coefficients between the sample face images and the variance of the coefficients between the sample makeup face images,
The selection means selects the plurality of evaluation basis vectors based on a difference between a coefficient variance between the sample face images and a coefficient variance between the sample makeup face images.
The makeup face image evaluation apparatus according to claim 1 or 2. The said evaluation information calculation means is a makeup | decoration face image evaluation apparatus of any one of Claim 1 to 3 which calculates the said evaluation information using the following evaluation function D (x).
x represents the evaluation image vector, x ₀ represents the target image vector, u _i represents the evaluation base vector, k represents the number of the evaluation base vectors, and T represents transposition. Based on the evaluation information calculated by the evaluation information calculation means, the proximity of the makeup face image to be evaluated to the target makeup face image is determined, and includes at least one of the determination result and the evaluation information The makeup face image evaluation apparatus according to any one of claims 1 to 4, further comprising output means for outputting result information in a predetermined form. By analyzing the independent components of a plurality of sets of sample face images and sample makeup face images each having a standardized shape, the coefficients of a plurality of base vectors are calculated for each sample face image and each sample makeup face image,
Based on the variation degree of the coefficient between the sample face images and the variation degree of the coefficient between the sample makeup face images in each of the basis vectors, a part of the plurality of basis vectors is used as a plurality of evaluation basis vectors. Selected,
Obtaining evaluation information of the evaluation target makeup face image from the inner product of the evaluation image vector obtained from the evaluation target makeup face image and the target image vector obtained from the target makeup face image and each of the evaluation base vectors;
A makeup face image evaluation method. Independent calculation of multiple base vector coefficients for each sample face image and each sample face image by independent component analysis of multiple sets of sample face images and sample face images that have been normalized in shape Component analysis means;
Based on the variation degree of the coefficient between the sample face images and the variation degree of the coefficient between the sample makeup face images in each of the basis vectors, a part of the plurality of basis vectors is used as a plurality of evaluation basis vectors. A selection means to select;
From the inner product of the evaluation image vector obtained from the makeup face image to be evaluated, the target image vector obtained from the target makeup face image, and the respective basis vectors for evaluation selected by the selection means, the makeup face image to be evaluated Evaluation information calculation means for acquiring evaluation information of
A program that makes a computer realize.