CN105761260A - Skin image affected part segmentation method - Google Patents

Abstract

The invention relates to a skin image affected part segmentation method comprising the steps that an input skin image is preprocessed, RGB component values are extracted, and the image is converted to an HSV color space; skin color areas and non-skin color areas are detected, and internal filling is performed in the skin color areas; pixel value histogram statistics is performed on the skin color area H component diagram obtained in 3), and a new skin color area H component diagram is obtained by using an Otsu threshold method; and binarization is performed on the skin color area H component diagram and morphological processing is performed. The skin image affected part segmentation method is suitable for segmentation of the affected part in multiple skin disease images and has the characteristics of high operation speed and high accuracy.

Description

A kind of skin image affected part dividing method

Technical field

The present invention relates to field of medical image processing, particularly to a kind of skin image affected part dividing method.

Background technology

Along with image processing techniques diagnoses, in the extensive use of medical domain, computer diagnosis and auxiliary, the hot topic becoming research.Dermatosis is the disease that a class is common, owing to its affected part can observe directly, so can realize dermopathic computer diagnosis and analysis by image processing techniques.The first step of dermatosis computer diagnosis is exactly split from the skin picture collected in region, dermatosis affected part, only achieves the accurate and efficient segmentation to affected part and could analyze the various features in affected part further.Current main difficulty has: (1), because skin is affected by various factors, dermopathic form and color distortion are bigger；(2) skin general red sector territory often occurs around affected part so that the interface edge of affected part and skin is also inconspicuous, by the partitioning algorithm effect not rationality based on gradient edge；(3) if the region outside there is skin in photo, this region is likely to segmentation is produced interference.

Summary of the invention

It is an object of the invention to provide a kind of speed of service fast, the skin image affected part dividing method that accuracy rate is high.Technical scheme is as follows:

1. a skin image affected part dividing method, comprises the following steps:

1) input skin image is carried out pretreatment, extract RGB component value, image is changed in HSV color space, and obtain the H component map of normalized value.

2) use the skin color detection algorithm based on rgb color space to detect area of skin color and non-area of skin color, area of skin color is carried out internal filling；

3) H-number by H component corresponding for non-area of skin color is zero, obtains the area of skin color H component map that skin is filled through inside；

4) to 3) in the area of skin color H component map that obtains carry out pixel value statistics with histogram, the point ordinate value making H component value be zero in rectangular histogram is equal to the average of front and back point ordinate value, by rectangular histogram to right translation 0.6, Otsu threshold method is used to calculate the optimal threshold T of H histogram of component；

5) to 3) in the H component value that obtains to left T+0.6, obtain new area of skin color H component map；

6) by 5) in the area of skin color H component map binaryzation that obtains, white portion is region, affected part, and black region is region, non-affected part, adopts the closed operation in Morphological scale-space eliminate white portion interior void and make edge-smoothing the image after binaryzation；

7) extract 6) in closed operation process after bianry image edge, marker edge in original image, and export image.

First input picture is transformed into HSV color space by rgb color space and obtains normalization H component by the present invention, then skin color detection algorithm is used to remove the interference of non-area of skin color, Otsu threshold method is used to calculate H histogram of component threshold value, H component is made to move to left the size of this threshold value, finally obtain segmentation result accurately by closing operation of mathematical morphology, the segmentation in the affected part suitable in multiple dermatosis image, has the speed of service fast, the feature that accuracy rate is high.

Accompanying drawing explanation

Fig. 1 is FB(flow block) of the present invention.

Fig. 2 is that the present invention is embodied as example figure.

Fig. 3 is H component exemplary plot.

Fig. 4 is Face Detection internal filling result exemplary plot.

Fig. 5 is that H component removes non-flesh tone portion exemplary plot.

Fig. 6 is exemplary plot after the translation of H component.

Fig. 7 is closing operation of mathematical morphology result exemplary plot.

Fig. 8 is affected part segmentation result output example image.

Detailed description of the invention

In order to further illustrate the present invention, provide an instantiation below in conjunction with the FB(flow block) of accompanying drawing 1 and the enforcement example figure of accompanying drawing 2-8.

Fig. 2 show a typical affected skin image.Wherein there is region, several piece affected part, in addition also have the background area outside two parts skin to interfere.Part affected part in figure and normal skin border are also inconspicuous.

Flow process shown in reference block Fig. 1, the skin image affected part cutting procedure that the present invention is concrete describes as follows:

Step one: the input picture shown in Fig. 2 is carried out noise suppression preprocessing.Image is transformed in HSV color space, obtains normalized H component map I₁, namely the value of H component is between 0 to 1.By R, G, B component calculate normalized H component formula be:

$H=\left\{\begin{array}{cc}0,& \begin{array}{cc}if& max=min\end{array}\\ \frac{1}{6}\&times;\frac{G-B}{\max -\min },& \begin{array}{cc}if& max=RandG\&GreaterEqual;B\end{array}\\ \frac{1}{6}\&times;\frac{G-B}{\max -\min }+1,& \begin{array}{cc}if& max=RandG<B\end{array}\\ \frac{1}{6}\&times;\frac{B-R}{\max -\min }+\frac{1}{3},& \begin{array}{cc}if& \max =G\end{array}\\ \frac{1}{6}\&times;\frac{R-G}{\max -\min }+\frac{2}{3},& \begin{array}{cc}if& \max =B\end{array}\end{array}\right.$

Wherein max=max{R, G, B}, min=min{R, G, B}.Fig. 3 show the H component map after normalization.

Step 2: use the skin color detection algorithm based on rgb color space to detect the area of skin color in image and non-area of skin color.Face Detection discrimination 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)]

Wherein y_j(1), y_j(2), y_j(3) R, G, B color component value of respectively each image pixel.The pixel meeting above-mentioned discriminant is judged to skin area, and in binary map, assignment is 1, is otherwise non-skin region, and in binary map, assignment is 0.Skin area inside is filled with, exports binary map I₂.Fig. 4 show the binary map after Face Detection and internal filling

Step 3: from I₂Statistical value is the position of 0, at I₁Middle correspondence position makes the H-number of this subregion be 0, obtains I₃, to remove the impact of non-area of skin color.Fig. 5 show I₃。

Step 4: to I₃In H component value carry out statistics with histogram, the some ordinate value making H component value be 0 in rectangular histogram is equal to the average of front and back point ordinate value.Due to the color character of the colour of skin and affected part, it is between 0-0.35 and 0.87-1 that the distribution of H histogram of component concentrates on value, when using Otsu threshold method to calculate threshold value, it is necessary to histogram distribution is adjusted.Consider that skin image is about 0.6 without pixel distribution at H component, by rectangular histogram to right translation 0.6.Otsu threshold method is used to calculate the optimal threshold T of H histogram of component.Namely the computational methods of threshold value T seek T value during following object function maximum.

${\mathrm{\&delta;}}^2\left(T\right)=\frac{{\&lsqb;\mathrm{\&mu;}w\left(T\right)-\mathrm{\&mu;}\left(T\right)\&rsqb;}^2}{w\left(T\right)\&lsqb;1-w\left(T\right)\&rsqb;}$

Wherein w (T) is H component value probability of occurrence between 0 to T, and μ (T) is threshold value is H component meansigma methods during T, and μ is the H component meansigma methods of general image.

Step 5: to I₃In H component value to left T+0.6, obtain new H component map I₄.Fig. 6 show the image after the translation of H component.

Step 6: by I₄Binaryzation obtains I₅, white portion is region, affected part, and black region is region, non-affected part.In order to eliminate white portion interior void and make edge-smoothing, the image after binaryzation is adopted the closed operation in Morphological scale-space, namely first expands post-etching:

Wherein B (x) is structural element.Finally obtain bianry image and be designated as I₆.Fig. 7 show the bianry image after Morphological scale-space.

Step 7: extract I₆Edge, mark edge in original image, and export image.Fig. 8 show output result images, and affected part segmenting edge uses green line labelling in the drawings.

Claims (1)

1. a skin image affected part dividing method, comprises the following steps:

1) input skin image is carried out pretreatment, extract RGB component value, image is changed in HSV color space, and obtain the H component map of normalized value.

2) use the skin color detection algorithm based on rgb color space to detect area of skin color and non-area of skin color, area of skin color is carried out internal filling；

3) H-number by H component corresponding for non-area of skin color is zero, obtains the area of skin color H component map that skin is filled through inside；

4) to 3) in the area of skin color H component map that obtains carry out pixel value statistics with histogram, the point ordinate value making H component value be zero in rectangular histogram is equal to the average of front and back point ordinate value, by rectangular histogram to right translation 0.6, Otsu threshold method is used to calculate the optimal threshold T of H histogram of component；

5) to 3) in the H component value that obtains to left T+0.6, obtain new area of skin color H component map；

6) by 5) in the area of skin color H component map binaryzation that obtains, white portion is region, affected part, and black region is region, non-affected part, adopts the closed operation in Morphological scale-space eliminate white portion interior void and make edge-smoothing the image after binaryzation；

7) extract 6) in closed operation process after bianry image edge, marker edge in original image, and export image.