JP6649749B2 - Image processing method for texture on skin surface and method for evaluating texture on skin surface using the same - Google Patents

Description

  The present invention relates to an image processing method for evaluating texture (texture) on the skin surface, and a method for evaluating skin texture using the image.

Visual information from the skin surface gives information on the shape in addition to color information such as spots and freckles. There are wrinkles and tarmi in the shape, but texture (texture) is also an important factor that expresses the homeostasis and beauty of the skin.
Since various factors are involved in the evaluation of texture, a skilled person generally judges it. However, a method of measuring the number of intersection points per unit area (Patent Document 1), thinning processing on an image, and the like are performed. A method of making a determination without human intervention, such as a method (Patent Document 2) and image processing by a short straight line matching method (Patent Document), is known.
In particular, recently, there has been a demand for the ability to quickly analyze the skin and the like of a consumer at a cosmetics sales site and to sell the most suitable cosmetics.

JP 2001-170028 A JP 2006-61170 A JP 2008-61892 A

  An object of the present invention is to provide an image processing method and an evaluation method for skin texture using the same, which do not require a skilled judge, accurately, anywhere, and can quickly evaluate the texture of the skin surface. .

As a result of extensive studies by the present inventors, as an image processing method,
This method solves this problem by extracting G (green) from an image file expressed in RGB of the skin surface, performing a rolling ball algorithm, histogram flattening on the extracted G (green) image, and then binarizing the image. It has been found that it is most suitable as a means of fixing.
After the binarization, a better result may be obtained by performing a median filter.
A scoring method was performed based on the result of performing the non-linear discriminant analysis on the image subjected to the image processing described above, thereby finding a method for evaluating the texture of the skin surface.

  More specifically, an image file expressed in RGB of the skin surface can be obtained by using a digital microscope or the like. USB Microscope M3, such as Scope R USB2.0, manufactured by Scalar Corporation, USB Microscope M3, Wireless Scope AirMicro A1 (U), Wireless Scope AirMicro A1, etc., Fortissimo Co., Ltd., etc.

When an image file expressed in RGB of the skin surface is obtained with the above-mentioned microscope or the like, G (green) is extracted from this image file.
A rolling ball algorithm and histogram flattening are performed on the extracted G (green) image.
The rolling ball algorithm is one method of background removal, and the radius of the rolling ball selects an optimum size according to the size of the image and the like.
Histogram flattening is also called equalization processing, and is processing for converting the distribution of histograms so as to be uniform.
After these processes, binarization is performed, and after binarization, noise removal is performed as necessary.
The binarization method includes a P-tile method, a modal method, a discriminant analysis method, and a minimum error method, which may be selected and used, but the discriminant analysis method (Otsu's method) is often used.
There are various methods for removing noise, and a moving average method, a median filter, and the like are often used. In the present invention, a median filter is most preferred.
The median filter is a process in which density values in an n × n local area are arranged in ascending order, and a density value in the middle is set as an output density of a pixel in the center of the area.

From the image obtained by the above-described method, it is possible to obtain various data such as the area of the skin ridge and its variation coefficient, the width and directionality of the skin sulcus.
The above processing can be carried out using publicly available software (ImageJ API / library; http://rsb.info.nih.gov/ij/, NIH, MD), including analysis macros Can be implemented automatically.

Furthermore, the present inventor was able to automatically classify the texture of the skin surface, which could only be performed by a skilled person, without human intervention, by non-linear discriminant analysis using the data obtained from the above images. .
The non-linear discriminant analysis is an algorithm for discriminating a data group into two or more groups (discrimination groups) to be discriminated. In the present invention, as an independent variable, “average area of a crust” and “a crust” When a scatter diagram of the subject's data is drawn using the “average area of the crust” and the “coefficient of variation of the crust area” using the “coefficient of variation of the area of An optimal function (non-linear discriminant function) to divide into is calculated. Then, when unknown data is input, it is possible to determine the texture state of the skin surface by determining which group group the unknown data belongs to with reference to the nonlinear discriminant function.
The classification of the discrimination group varies depending on the purpose, but is usually 3 to 20, and is often used in about 4 to 10 in many cases.
For the non-linear discriminant analysis, SAS (SAS), Stata (Lightstone), SPSS (IBM), S_PLUS (Mathematical Systems) and the like can be used, and the analysis can be performed using free software "R".

Next, an example will be described.
1-1. A JPEG image file was obtained on the skin surface of the female cheek with a digital videoscope. (Conducted by 78 people)
Using ImageJ, the following operation was performed on the image file expressed in RGB.
1-2. An image file expressed in RGB was decomposed into RGB, and a G (green) image was extracted.
1-3. A rolling ball algorithm was performed on the extracted G (green) image.
1-4. The histogram was flattened.
1-5. It was binarized by Otsu's method.
1-6. A median filter (average value of 3 × 3 pixels) was performed.
By the above method, an image capable of evaluating the texture of the skin surface was obtained.

2-1. From this image, the average value and the coefficient of variation of the area of the crust were obtained.
Non-linear discriminant analysis was performed by software “R” using this numerical value and a score described later as teacher data.
An image obtained by a digital videoscope before being operated with ImageJ was classified by an expert into four classes from a condition with almost no texture (score 1) to an ideal condition with a texture (score 4). .

As a result of the nonlinear discriminant analysis, an optimal nonlinear discriminant function for dividing into four discriminant groups was obtained.
For confirmation, the above-mentioned 78 data were re-classified into four classes using the nonlinear discriminant function obtained above, and as a result, as shown in FIG. 3, the coincidence rate with the visual score by an expert was 69%. There were no results that differed by more than two stages between the two, indicating that the two could be used sufficiently as a means to replace visual observation by a skilled person.

Image A = Image obtained with a digital video scope Image B = Image of G (green) extracted from Image A Image C = Image obtained by performing rolling ball algorithm (radius 50 pixels) on Image B Image D = Image C subjected to histogram flattening processing Image E = Image D obtained by binarizing image D using the Otsu method Image F = Image obtained by applying a median filter to image E Representative examples of visual scores by skilled workers and images processed by the examples Comparison between the visual score of the expert and the result of the nonlinear discriminant analysis by the examples

Claims (4)

G (green) is extracted from the image file of the skin surface expressed in RGB, and the extracted G (green) image is subjected to a rolling ball algorithm, histogram flattening, and then binarized image processing of the skin surface texture Method.   2. The image processing method according to claim 1, wherein a median filter is performed after binarization.   A method for evaluating texture on the skin surface, characterized in that an image processed by the method according to any one of claims 1 and 2 is scored based on a result of performing a nonlinear discriminant analysis. 3. The image processing method according to claim 1, wherein the image file is an image file obtained by using a digital microscope.