JP4285037B2 - Method for forming simulated skin image - Google Patents

Description

【００３６】
表１及び図７から、５階層の再合成画像も２階層の再合成画像も、自然なテクスチャを有し、色ズレもなかったことから、本発明によれは、自然なテクスチャの肌画像を形成できることがわかる。さらに、５階層の再合成画像は２階層の再合成画像よりもシミが少なく、２０代の女性の当初の画像と同様のテクスチャとなっていたことから、５階層の再合成画像は、５０代の女性の画像からシミを目立たなくしたシミュレーション画像となることがわかる。
【００３７】
【発明の効果】
本発明によれば、より自然なテクスチャの肌の画像を形成するために有用な画像の解析方法が提供され、また、色ズレを生じさせることなく、より自然なテクスチャを有する肌のシミュレーション画像を形成することが可能となる。したがって、例えば、シミを大きく、濃く、あるいは小さく、薄くしたり、色ムラのパターンを自然に変化させることが可能となる。
【図面の簡単な説明】
【図１】 肌のシミュレーション画像の形成方法の工程図である。
【図２】 肌のシミュレーション画像の形成システムの模式図である。
【図３】 独立成分分析の説明図である。
【図４】 ヒストグラムの説明図である。
【図５】 ヒストグラムのマッチングの説明図である。
【図６】 実施例で形成した５０代女性と２０代女性と再合成画像のヒストグラムである。
【図７】 ５０代の女性と２０代の女性のそれぞれの内部反射光画像と５階層の再合成画像である。
【符号の説明】
１ 光源
２、３ 偏光板
４ デジタルカメラ
５ パーソナルコンピュータ
６ ディスプレイ
７ プリンタ
Ｐ 被験者[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a skin texture analysis method, a skin simulation image forming method using the analysis method, and a system therefor.
[0002]
[Prior art]
It is known that human skin color depends on melanin content and hemoglobin content. For example, brown skin has a high melanin content and white skin has a low hemoglobin content. Ruddy skin has a high hemoglobin content, particularly a high oxygen partial pressure hemoglobin content, and poor ruddy skin has a low hemoglobin content. In addition, local melanin content and hemoglobin content affect the appearance of spots and acne.
[0003]
As a method for measuring the melanin and hemoglobin components that contribute to the skin color, as the spectral image data of the reflected light from the skin, the internal reflected light component of the skin is detected using polarized light, and component analysis is performed on the internal reflected light component. A method has been proposed in which the amount of melanin component and the amount of hemoglobin component in skin color are obtained (see Patent Document 1).
[0004]
On the other hand, with the progress of computer graphic technology, it is possible to reproduce human skin on an image with a more natural texture than before, taking into consideration the layer structure of the skin and the reflection characteristics based on it.
[0005]
However, the actual situation is that reproduction of the spatial distribution of skin, such as spots and acne, found in actual skin is performed manually. That is, in the formation of a simulation image of the skin, color adjustment such as making the skin color brown or lightening as a whole according to the information on the melanin content and hemoglobin content of the skin is automatically performed by image processing using a computer. However, spots and the like must be manually distributed on the image.
[0006]
[Patent Document 1]
JP-A-2002-200050 gazette
[Problems to be solved by the invention]
Therefore, the present invention provides a skin image analysis method useful for forming an image of a skin having a texture that gives a more natural impression, and a method for forming a simulation image of a skin having a texture that gives a more natural impression, and therefore It aims at providing the system used for.
[0008]
[Means for Solving the Problems]
The present inventor forms an isolated image of skin pigment components such as melanin and hemoglobin by performing independent component analysis on the internally reflected light image of the skin formed using polarized light. By applying the pyramid method to obtain images of multiple layers with different spatial frequencies, and obtaining a histogram of pixel intensity for each image of each layer, the distribution of this histogram becomes an effective parameter of the skin texture, By obtaining this histogram for a specific target skin, matching the histogram of the subject's image with the histogram of the target skin, and recombining the images of each layer, the texture of the subject's image is totally misaligned. It was found that a simulation image close to the texture of the target skin can be formed without causing it to occur.
[0009]
That is, the present invention forms an internally reflected light image of the skin using polarized light,
By performing independent component analysis on the internally reflected light image, a component image of at least one pigment component constituting the skin color is extracted,
By applying an image pyramid method to the dye component image, a plurality of images having different spatial frequencies are formed,
A skin texture analysis method for obtaining a histogram of pixel intensities for images having different spatial frequencies is provided.
[0010]
In addition, the present invention provides a method for forming a simulation image of skin using the analysis method described above.
Form an internal reflection image of the subject's skin using polarized light,
By performing independent component analysis on the internally reflected light image, a component image of at least one pigment component constituting the skin color is extracted,
By applying an image pyramid method to the dye component image, multiple layers of images having different spatial frequencies are formed,
Obtain pixel intensity histograms for images with different spatial frequencies,
A simulation image in which the histogram is matched with a histogram of a specific skin acquired by the same method, the skin image of the subject is re-synthesized, and the skin texture in the subject image is close to the specific skin texture Provide a method of forming.
[0011]
Furthermore, the present invention provides, as a system for forming this simulation image, an image forming means capable of forming an internal reflected light image of skin using polarized light,
A component image of at least one pigment component constituting the skin color is extracted by performing independent component analysis on the color of the internally reflected light image, and an image pyramid method is applied to the pigment component image, so that a plurality of different spatial frequencies can be obtained. Forming an image, obtaining a histogram of pixel intensities for images with different spatial frequencies, matching the histogram against a histogram of a specific skin acquired in a similar manner, and re-synthesize the skin image of the subject, Provided is a calculation means for forming a simulation image in which a skin texture is close to the specific skin texture in a subject image, and a skin simulation image forming system including a display for displaying the simulation image.
[0012]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, the present invention will be described in detail with reference to the drawings. In each figure, the same numerals indicate the same or equivalent components.
[0013]
FIG. 1 is a process diagram of an embodiment of a method for forming a simulation image of skin according to the present invention, and FIG. 2 is a schematic diagram of a system according to an embodiment of the present invention for forming the simulation image. FIG. 1 includes the skin texture analysis method of the present invention, and FIG. 2 is also a system used to analyze the skin texture.
[0014]
This system includes a light source 1 that illuminates the subject P, a digital camera 4 that captures an image of the subject P, a polarizing plate 2 and 3 that are detachably provided on the front surface of the light source 1 and the digital camera 4, respectively, and various operations that will be described later. A personal computer 5 having a function, a display 6 connected to the personal computer 5 and capable of switching or simultaneously displaying a normal image, an internal reflection light image, a surface reflection light image, a simulation image, etc. of the skin of the subject P; A printer 7 is provided.
[0015]
In this system, the simulation image is formed by first forming an internally reflected light image of the face image of the subject P using the polarizing plates 2 and 3 and the digital camera 4. Here, the internally reflected light image is obtained by attaching the polarizing plate 3 on the front surface of the digital camera 4 so that the polarization direction is orthogonal to the polarizing plate 2 on the front surface of the light source 1 and removing the surface reflected light component. Form. In this case, in order to obtain an internally reflected light image in which shading is canceled as much as possible, it is preferable to arrange a plurality of light sources 1 (for example, from four directions, up, down, left, and right) so as to illuminate the face of the subject P from all around. .
[0016]
If a surface reflected light image is obtained together with the internally reflected light image, a more natural simulation image is obtained by recombining the internally reflected light image obtained by simulation and the surface reflected light image. be able to. Here, the surface reflected light image is obtained by polarizing the polarizing plate 3 on the front surface of the digital camera 4 with the polarizing plate 3 mounted so that the polarizing direction of the polarizing plate 2 on the front surface of the light source 1 is the same. It is preferably formed from the difference between the plate 2 and the image obtained by mounting so that the polarization directions are orthogonal. In order to obtain a surface reflected light image in which the shadow is emphasized, the light source 1 preferably illuminates the face of the subject P from one direction (for example, only from above the face).
[0017]
For reference, it is preferable to take a picture without using the polarizing plates 2 and 3 and obtain a normal image.
[0018]
Next, an independent component analysis is performed on the internally reflected light image by a calculation unit incorporated in the personal computer 5 to extract component images of the melanin component and the hemoglobin component constituting the skin color.
[0019]
Here, independent component analysis refers to the skin layer structure, the epidermal layer containing melanin as the main pigment component, the dermis layer containing hemoglobin as the main pigment component, and the subcutaneous tissue containing other pigment components. This is an analysis method that separates and extracts the signals of each layer from the image signal, considering that the signal is emitted independently from each layer and the mixture of them is an image signal.
[0020]
More specifically, for RGB of the image signal, -log (R), -log (G), and -log (B) are assigned to the x-axis, y-axis, and z-axis, respectively, and skin color is mapped to the color space there. As shown in FIG. 3 (a), it is found that two components contribute to the skin color because it is distributed in a substantially planar shape. This independent two-component signal intensity is considered to correspond to the amount of melanin or hemoglobin, respectively, and the skin color is closer to the component vector of melanin (-log (B), as shown in Fig. 3 (b). ) And the hemoglobin component vector (the one closer to -log (G)). Therefore, a signal representing the amount of melanin or a signal representing the amount of hemoglobin is extracted from the signal of the internally reflected light image of each subject, and a distribution image of the melanin component amount and a distribution image of the hemoglobin component amount are output.
[0021]
Details of such analysis processing and image processing are described in Vol. 16, No. 9 / September 1999 / J. Opt. Soc. Am. A 2169, and commercially available image analysis software (for example, This can be done by installing Adobe Photoshop).
[0022]
Next, an image pyramid method is applied to each dye component image extracted from the internally reflected light image to form multiple layers of images with different spatial frequencies (DJHeeger, JRBergen, COMPUTER GRAPHICS PROCEEDINGS, p229-238). (1995).
[0023]
The image pyramid method is a type of subband transformation, which decomposes an image into a set of subbands with different resolutions from high resolution to low resolution, and processes each subband independently of the other subbands. Alternatively, it is a method of recombining the decomposed subbands. There are various types of image pyramids depending on their basis functions, such as Gaussian pyramid, Laplacian pyramid, steerable pyramid, etc. It is preferable to use a Laplacian pyramid that holds square frequency information.
[0024]
Images of a plurality of hierarchies obtained by the image pyramid method hold spatial frequency information related to texture structure identification. Therefore, in the present invention, a histogram of a plurality of layers of the pigment component image of the subject (with the pixel intensity as the horizontal axis, for any pixel intensity interval, the probability of dividing the number of pixels corresponding to that interval by the total number of pixels) Distribution on the vertical axis) is used as an evaluation parameter for the skin texture. For example, when the distribution of the histogram obtained by applying the image pyramid method to the melanin component image is widened as shown in FIG. 4A, the distribution is narrower than when it is narrowed as shown in FIG. When the peak of the histogram is shifted to the higher pixel intensity as shown in FIG. 6C, the color unevenness is large, as shown in FIG. This means that the color of the melanin pigment component in the image is lighter than when the peak of the histogram is shifted to the smaller pixel intensity.
[0025]
Such a histogram is obtained for each layer image obtained by the image pyramid method. In this case, if the number of layers is too small, the accuracy of analysis is insufficient, and if it is too large, the calculation time becomes long. It is preferable to set it to ~ 7.
[0026]
In the present invention, the histogram of each layer of the image pyramid is matched (approximation processing) with the histogram of the target image of a specific texture, and the image of each layer is recombined. Thereby, it is possible to form a simulation image in which only the texture is brought close to the texture of the target image without causing color shift in the subject's image.
[0027]
More specifically, histogram matching is performed by first accumulating the probability of the histogram of the subject image (pixel strength vs probability) and the inverse cumulative distribution function (probability vs pixel strength) of the target image of a specific texture. ) And using these, the process of replacing the pixel value of the subject image with the pixel value of the target image is repeated for all the pixels. For example, as shown in FIG. 5A, the cumulative distribution function of the subject's image, as shown in FIG. 5B, the cumulative distribution function of the target image, and as shown in FIG. Find the inverse cumulative distribution function of. Then, from the figure (a), the pixel having a pixel intensity of 30 in the subject's image is calculated from the cumulative distribution function to have a probability of 0.32, and then the inverse cumulative distribution function is used to calculate a pixel intensity of 20 having a probability of 0.32. Is calculated and replaced with the pixel value of the subject's image. Such processing is performed on all pixels.
[0028]
In the simulation image forming method of the present invention, it is preferable that the application of the image pyramid method and the subsequent matching of the histogram are performed for at least both the component image of the melanin pigment and the component image of the hemoglobin pigment.
[0029]
After matching the histograms, re-synthesize the images in each layer by adding them to the image on the next layer in order from the lower layer, obtain a re-synthesized melanin image and a re-synthesized hemoglobin image, and then synthesize them Thus, a simulation image of the internally reflected light image can be obtained. Furthermore, by recombining the simulation image of the internally reflected light image thus obtained and the above-mentioned surface reflected light image of the subject, the texture is brought closer to the target image with respect to a normal image taken without using polarized light. A more natural simulation image can be obtained.
[0030]
In forming a simulation image as described above, a target image to be matched (approximation processing) may be that of a specific person, or an average face image of a plurality of persons may be used as a virtual model. . In addition, a database that associates subjective evaluations of texture (with youthful tension, few spots, smoothness, etc.) and histograms for various images in advance and subjective evaluation desired by the subject , A texture image corresponding to the subjective evaluation may be selected using the database.
[0031]
The skin texture analysis method of the present invention is, for example, an evaluation of the degree of sunburn, how the suntanned skin returns to the original skin, the effect of the whitening cosmetic on the stain, the change of the skin texture due to the application of the foundation, etc. Can be used. The skin simulation image forming method of the present invention is useful for viewing changes in skin texture without causing color shift in the image. For example, it is possible to form a simulation image of a young man with a smooth tension that has few spots, small spots, or lighter spots, without changing the color tone of the skin other than the spots from middle-aged skin with many spots.
[0032]
【Example】
Example 1
Females in their 50s and 20s were used as subjects, and internal reflection light images of each skin were obtained using polarized light, and independent image analysis was performed on these images to obtain melanin component images and hemoglobin component images. By applying the Laplacian pyramid method, images of 5 layers were obtained, and histograms were obtained for each layer. Then, the histogram of each layer of women in their 50s is matched with the histogram of the corresponding layer of women in their 20s, and the melanin component image and the hemoglobin component image of each woman in their 50s are respectively matched using the histogram of each layer after matching. They were re-synthesized and further synthesized. FIG. 6 shows, for a melanin component image, a five-layer histogram of a woman in her 50s, a five-layer histogram of a woman in her twenties, and a histogram of a five-layer image that has been matched.
[0033]
A synthetic image was obtained by repeating the above-described method except that a two-layer image was formed instead of forming a five-layer image by the Laplacian pyramid method.
[0034]
The original internal reflection image of a woman in her 50s and a woman in her 20s, a composite image obtained by recombining images of five layers, and a composite image obtained by recombining images of two layers, The appearance of the spots was visually observed. The results are shown in Table 1. Further, FIG. 7 shows a composite image using the initial internally reflected light images and the five-layer recombined images of the women in their 50s and 20s.
[0035]
[Table 1]

[0036]
From Table 1 and FIG. 7, since the re-synthesized image of 5 layers and the re-synthesized image of 2 layers had a natural texture and there was no color shift, according to the present invention, a skin image having a natural texture was obtained. It can be seen that it can be formed. Furthermore, the 5th layer re-synthesized image has fewer spots than the 2nd layer re-synthesized image and has the same texture as the original image of women in their 20s. It turns out that it becomes a simulation image which made the stain inconspicuous from the image of women.
[0037]
【The invention's effect】
According to the present invention, an image analysis method useful for forming an image of a skin having a more natural texture is provided, and a simulation image of a skin having a more natural texture can be obtained without causing a color shift. It becomes possible to form. Therefore, for example, it is possible to make the stain large, dark, small, thin, or to change the color unevenness pattern naturally.
[Brief description of the drawings]
FIG. 1 is a process diagram of a method for forming a simulation image of skin.
FIG. 2 is a schematic diagram of a system for forming a simulation image of skin.
FIG. 3 is an explanatory diagram of independent component analysis.
FIG. 4 is an explanatory diagram of a histogram.
FIG. 5 is an explanatory diagram of histogram matching.
FIG. 6 is a histogram of a recombined image of a woman in her 50s and a woman in her 20s formed in the example.
FIG. 7 is an internally reflected light image and a five-layer recomposed image of women in their 50s and 20s.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Light source 2, 3 Polarizing plate 4 Digital camera 5 Personal computer 6 Display 7 Printer P Subject

Claims (4)

Form an internal reflection image of the subject's skin using polarized light,
By performing independent component analysis on the internally reflected light image, a component image of at least one pigment component constituting the skin color is extracted,
By applying an image pyramid method to the dye component image, multiple layers of images having different spatial frequencies are formed,
Obtain pixel intensity histograms for images with different spatial frequencies,
A simulation image in which the histogram is matched with a histogram of a specific skin acquired by the same method, the skin image of the subject is re-synthesized, and the skin texture in the subject image is close to the specific skin texture How to form. How the dye component forms a skin simulation image according to claim 1, wherein the melanin component or hemoglobin component. Image forming means capable of forming an internally reflected light image of the skin using polarized light;
A component image of at least one pigment component constituting the skin color is extracted by performing independent component analysis on the color of the internally reflected light image, and an image pyramid method is applied to the pigment component image, so that a plurality of different spatial frequencies can be obtained. Forming an image, obtaining a histogram of pixel intensities for images with different spatial frequencies, matching the histogram against a histogram of a specific skin acquired in a similar manner, re-synthesizing the skin image of the subject, A calculation system for forming a simulation image in which a skin texture is close to the specific skin texture in a subject image, and a skin simulation image forming system including a display for displaying the simulation image. The skin simulation image forming system according to claim 3 , further comprising: a database in which subjective evaluation of texture and the histogram are associated with each other for a plurality of skin images, and a histogram as a matching model can be extracted from the database.

Images