CN105761260B - A kind of skin image affected part dividing method - Google Patents

Abstract

The invention relates to a skin image lesion segmentation method, comprising: preprocessing an input skin image, extracting RGB component values, converting the image into an HSV color space; detecting skin-colored areas and non-skinned areas, and filling the skin-colored area; ) to perform pixel value histogram statistics on the H component map of the skin color region, and use the Otsu threshold method to obtain a new H component map of the skin color region; binarize the H component map of the skin color region, and perform morphological processing. The invention is applicable to the segmentation of affected parts in various skin disease images, and has the characteristics of fast running speed and high accuracy.

Description

A skin image lesion segmentation method

technical field

The invention relates to the field of medical image processing, in particular to a skin image lesion segmentation method.

Background technique

With the wide application of image processing technology in the medical field, computer diagnosis and assisted diagnosis have become a hot topic in research. Skin disease is a kind of common disease. Since the affected area can be directly observed, computer diagnosis and analysis of skin disease can be realized through image processing technology. The first step in the computer diagnosis of skin diseases is to segment the affected area of the skin disease from the collected skin pictures. Only when the accurate and efficient segmentation of the affected area is realized can the various characteristics of the affected area be further analyzed. The main difficulties at present are: (1) Because the skin is affected by many factors, the shape and color of skin diseases are quite different; (2) There are often red areas around the affected area, so that the border between the affected area and the skin is not obvious , the effect of segmentation algorithm based on gradient edge is not rational; (3) If there is an area other than the skin in the photo, this area may interfere with the segmentation.

Contents of the invention

The purpose of the present invention is to provide a skin image lesion segmentation method with high speed and high accuracy. Technical scheme of the present invention is as follows:

1. A skin image lesion segmentation method, comprising the following steps:

1) Preprocess the input skin image, extract the RGB component value, convert the image into the HSV color space, and obtain the H component map of the normalized value.

2) Use a skin color detection algorithm based on the RGB color space to detect skin color areas and non-skin color areas, and fill the skin color area internally;

3) Set the H value of the H component corresponding to the non-skin color area to zero, and obtain the H component map of the skin color area filled with the skin;

4) Perform pixel value histogram statistics on the H component map of the skin color area obtained in 3), the ordinate value of the point whose H component value is zero in the histogram is equal to the mean value of the ordinate values of the front and rear points, and shift the histogram to the right by 0.6 , using the Otsu threshold method to calculate the optimal threshold T of the H component histogram;

5) Translate the H component value obtained in 3) to the left by T+0.6 to obtain a new H component map of the skin color area;

6) Binarize the H-component map of the skin color area obtained in 5), the white area is the affected area, and the black area is the non-affected area, and the closed operation in the morphological processing is used to eliminate the inner cavity of the white area for the binarized image and smooth the edges;

7) Extract the edge of the binary image processed by the closed operation in 6), mark the edge in the original image, and output the image.

The present invention first converts the input picture from the RGB color space to the HSV color space to obtain the normalized H component, then uses the skin color detection algorithm to remove the non-skin color area interference, and uses the Otsu threshold method to calculate the H component histogram threshold, so that the H component is shifted to the left. The size of the threshold value, and finally use the morphological closed operation to obtain accurate segmentation results, which is suitable for the segmentation of affected areas in various skin disease images, and has the characteristics of fast running speed and high accuracy.

Description of drawings

Fig. 1 is a flow chart of the present invention.

Fig. 2 is a diagram for a specific implementation example of the present invention.

Figure 3 is an example diagram of the H component.

Figure 4 is an example of skin color detection and internal filling results.

Fig. 5 is an example diagram of H component removal of non-skin color parts.

Fig. 6 is an example diagram after translation of the H component.

Figure 7 is an example diagram of the results of the morphological closing operation.

Figure 8 is an example image of the output of the lesion segmentation result.

Detailed ways

In order to further illustrate the present invention, a specific example is given below in conjunction with the flow chart of accompanying drawing 1 and the implementation examples of accompanying drawing 2-8.

Figure 2 shows a typical image of diseased skin. There are several affected areas, in addition to two background areas outside the skin causing interference. In the picture, the boundary between some affected areas and normal skin is not obvious.

With reference to the process shown in block diagram 1, the specific skin image lesion segmentation process of the present invention is described as follows:

Step 1: Perform denoising preprocessing on the input image shown in Fig. 2 . Convert the image to the HSV color space to obtain a normalized H component map I ₁ , that is, the value of the H component is between 0 and 1. The formula for calculating the normalized H component from the R, G, and B components is:

where max=max{R,G,B}, min=min{R,G,B}. Figure 3 shows the normalized H component map.

Step 2: Use a skin color detection algorithm based on RGB color space to detect skin color areas and non-skin color areas in the image. The skin color detection discriminant formula is as follows:

[y _j (1)>95&&y _j (2)>40&&y _j (3)>20&&y _j (1)-y _j (3)>15&&y _j (1)-y _j (2)>15]

||[y _j (1)>200&&y _j (2)>210&&y _j (1)>170&&abs(y _j (1)-y _j (3)]

<=[15&&y _j (1)>y _j (3)&&y _j (2)>y _j (3)]

Among them, y _j (1), y _j (2), and y _j (3) are the R, G, and B color component values of each image pixel, respectively. Pixels that meet the above discriminant formula are judged as skin areas, and are assigned a value of 1 in the binary image; otherwise, they are non-skin areas, and are assigned a value of 0 in the binary image. Fill the inside of the skin area and output the binary image I ₂ . Figure 4 shows the binary image after skin color detection and internal filling

Step 3: From the position where the statistical value of I ₂ is 0, set the H value of this part of the area to 0 at the corresponding position in I ₁ to obtain I ₃ to remove the influence of the non-skinned area. Figure 5 shows I ₃ .

Step 4: Perform histogram statistics on the H component value in I ₃ , the ordinate value of the point whose H component value is 0 in the histogram is equal to the mean value of the ordinate values of the front and rear points. Due to the skin color and the color characteristics of the affected area, the histogram distribution of the H component is concentrated between 0-0.35 and 0.87-1. When using the Otsu threshold method to calculate the threshold, the histogram distribution needs to be adjusted. Considering that the skin image has no pixel distribution around the H component of 0.6, shift the histogram to the right by 0.6. The optimal threshold T for the H component histogram was calculated using the Otsu threshold method. The calculation method of the threshold T is to find the T value at the maximum of the following objective function.

Where w(T) is the occurrence probability of the H component value between 0 and T, μ(T) is the average value of the H component when the threshold is T, and μ is the average value of the H component of the overall image.

Step 5: Translate the H component value in I ₃ to the left by T+0.6 to obtain a new H component map I ₄ . Figure 6 shows the image after H component translation.

Step 6: Binarize I ₄ to obtain I ₅ , the white area is the affected area, and the black area is the non-affected area. In order to eliminate the internal cavity of the white area and smooth the edge, the closed operation in the morphological processing is used for the binarized image, that is, first dilated and then corroded:

where B(x) is a structural element. The finally obtained binary image is denoted as I ₆ . Figure 7 shows the binary image after morphological processing.

Step 7: Extract the edge of I ₆ , mark the edge in the original image, and output the image. Figure 8 shows the output result image, and the segmentation edge of the lesion is marked with a green line in the figure.

Claims (1)

1. A skin image lesion segmentation method, comprising the following steps: 1) Preprocess the input skin image, extract the RGB component value, convert the image into the HSV color space, and obtain the H component map of the normalized value; 2) Use a skin color detection algorithm based on the RGB color space to detect skin color areas and non-skin color areas, and fill the skin color area internally; 3) Set the H value of the H component corresponding to the non-skin color area to zero to obtain the H component map of the skin color area filled inside; 4) Perform pixel value histogram statistics on the H component map of the skin color area obtained in 3), the ordinate value of the point whose H component value is zero in the histogram is equal to the mean value of the ordinate values of the front and rear points, and shift the histogram to the right by 0.6 , using the Otsu threshold method to calculate the optimal threshold T of the H component histogram; 5) Translate the H component value obtained in 3) to the left by T+0.6 to obtain a new H component map of the skin color area; 6) Binarize the H-component map of the skin color area obtained in 5), the white area is the affected area, and the black area is the non-affected area, and the closed operation in the morphological processing is used to eliminate the inner cavity of the white area for the binarized image and smooth the edges; 7) Extract the edge of the binary image processed by the closed operation in 6), mark the edge in the original image, and output the image.