CN113393458A - Hand wound detection method based on wound weighting significance algorithm - Google Patents

Abstract

In this paper, a method for hand wound detection based on wound weighted saliency is proposed. In general, the salient effect of saliency algorithms on images is relatively insignificant for the extraction of specialized regions, and has many limitations. This method mainly uses GrabCut for foreground extraction, and then performs initial quantization, advanced color screening, conversion color space and GB segmentation on the hand image respectively, and the influence of the distance between regions can be calculated in different segmentation regions in Lab space. , and the weighting of the wound color feature is introduced to make the wound more prominent. In addition, the moving visual focus is also introduced, the real visual focus in the image is calculated and the saliency value is recalculated to obtain the final saliency map of the prominent wound. A large number of experiments have proved that this method is very effective for wound detection.

Description

A hand wound detection method based on wound weighted saliency algorithm

technical field

The invention mainly relates to image processing technology, in particular to a hand wound detection method based on wound weighted significance

Background technique

In recent years, with the emergence of concepts such as "smart city" and "smart kitchen", on the one hand, it reflects that people's demand for scientific and technological progress is greater than before; on the other hand, the research on the hand also has a new development process. . Nowadays, the construction of smart kitchen ^[4] is also a hot topic, because people are eager to integrate the latest technology into their daily life and hope that their life will be "high-tech". The design concept and materialization implementation process of smart kitchen includes five principles, namely, intelligent kitchen facilities, humanized operation, low carbonization, open communication platform, and integration of dining and kitchen. The most important facility for facility intelligence may be the morning detector. The morning detector is an intelligent device that detects whether the chef's hand is qualified. The detection algorithm of the hand content and the morning detector can be combined to achieve its purpose. However, throughout the country and abroad, the research on the hand is nothing more than the recognition of gestures and the detection of key points, and there is very little research on the content of our hands. Therefore, in this relatively embarrassing situation, the detection technology of hand color image content is of great significance.

A very important content detection in hand color image content is the detection of hand wounds. For the color image of the hand, if there is a wound on the hand, whether it is a fresh wound or a scabbed wound, it is more conspicuous than the whole hand image, and it is more conspicuous within the normal vision range of people. Therefore, the saliency algorithm is mainly used in the wound detection part of the color image of the hand, and the regional saliency algorithm is applied to the color image extracted from the hand first.

The more famous saliency algorithms are the Itti algorithm based on the fusion feature integration theory of biological models, the central-peripheral operator visual saliency detection method FT algorithm based on color features, etc. There are also RC algorithms based on global area contrast that have introduced gravity in recent years. The model detects the MRC algorithm of the most salient part of the human eye, etc., but these algorithms do not have a good effect on wound detection in hand images, so a DRC saliency algorithm is proposed based on the MRC algorithm.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose a hand wound detection method based on wound weighted saliency, so as to achieve relatively accurate highlighting of the wound area of the hand in the hand image with complex background. In this invention, the realization process is automatically completed by the computer, and the user only needs to input the hand image, and then he can detect the wound area in the image by himself.

The technical scheme of the present invention is as follows:

Step S1, use the GrabCut technology to extract the foreground of the original image, and obtain the segmentation map K extracted to the hand;

In step S2, initial quantization and advanced color screening are performed on the segmentation map K obtained in step S1, to obtain an image set K ² with small visual differences after quantization but with a reduced number of colors;

Advanced color screening is actually a further image quantization, the main purpose is to ensure that 95% of the original image color and high quality are preserved;

At the same time, color space conversion is performed on the obtained image set K ² to obtain an image set K ³ ;

If you want to convert from the most common RGB color space to the Lab color space, you need to go through the XYZ color space in the middle. The purpose of converting the color space is to be more in line with human vision;

In step S3, the segmentation map K obtained in step S1 is segmented using GB technology to obtain a color segmentation map C ^m ;

GB segmentation technology is the most commonly used image segmentation technology. It mainly uses the similarity between pixels and pixels, and converts the image into a map for region division. In the process of segmentation, different regions can be divided into different regions. Coloring in different colors to better display the segmentation effect;

In step S4, ^on the Lab color space image K3 obtained in step S2, the color distance between regions is calculated according to the color segmentation map ^Cm obtained in step S3, and wound feature weighting is introduced to obtain an initial saliency map ^K4 ;

The main principle used to calculate the initial significant value is that the farther the regions are, the less they affect each other, and the closer they are, the greater the impact;

Wound weighting mainly calculates the similarity between the salient area of the hand image and the standard wound area. The more similar the area is, the closer the area is to the wound, thus increasing the weight value and highlighting the wound area.

In step S5, the initial saliency map ^K4 obtained in step S4 is recalculated for the visual focus of the image and the real hand wound saliency map ^K5 .

The initial visual focus is usually the center of the image, because the center of gravity of most images does not coincide with the center. In order to calculate the saliency value more accurately, the saliency gravity model is introduced here to calculate the final visual focus, mainly calculating the direction and distance of the point. Get the ultimate visual focus.

Description of drawings

Various aspects of the present invention will be more clearly understood by the reader after reading the detailed description of the invention with reference to the accompanying drawings. in,

Fig. 1 is the flow chart of the DRC saliency algorithm based on the MRC saliency algorithm of the present invention;

Fig. 2 is a specific implementation process diagram of the present invention;

Figure 3 is a comparison chart of wound detection between the final and other methods.

Detailed ways

Step S1, use the GrabCut technology to extract the foreground of the original image, and obtain the segmented image extracted to the hand;

In step S2, initial quantization is performed on the segmentation image obtained in step S1, and the quantization process is to quantize the color value of each channel in the image into 12 different values in the RGB color space, and the initial quantization operator is: P:R ^{m ×n} →R ^m×n , for each pixel of each image, the quantized color value is:

分别表示原始图像中像素点在RGB空间每个通道的颜色值。

分别表示量化后图像中每个像素点在RGB空间的颜色通道值；Int(.)表示浮点型转换为整型。Among them, s=1,2,...,m; t=1,2,...,n,

Respectively represent the color value of each channel of the pixel in the original image in the RGB space.

Respectively represent the color channel value of each pixel in the RGB space after quantization; Int(.) represents the conversion of floating point type to integer type.

量化函数如下：The image after primary quantization cannot meet our requirements. In order to ensure that the quantized image contains 95% of the pixels of the original image and does not affect the quality, advanced color screening is also required, that is, further quantization. for every pixel

The quantization function is as follows:

分别表示量化后图像中像素点在RGB空间的每个通道的颜色值。in,

Respectively represent the color value of each channel in the RGB space of the pixel in the quantized image.

The number of colors in the quantized image set is greatly reduced, which reduces the calculation time and speeds up the speed.

At the same time, to calculate the salient value of the image, it is necessary to convert the color space of the image first. The conversion process is to convert the image from the RGB color space to the XYZ color space first. The conversion formula is:

分别表示的是经过量化后得到的图像中像素点

在RGB颜色空间下三个颜色通道(RGB)的颜色值，

分别表示的是经过以上公式转换得到的图像中像素点H_s,t在XYZ颜色空间下三个颜色通道(XYZ)的颜色值。并且B为转换矩阵，定义如下：in,

Respectively represent the pixels in the image obtained after quantization

The color values of the three color channels (RGB) in the RGB color space,

Respectively represent the color values of the three color channels (XYZ) of the pixel point H _s,t in the XYZ color space in the image converted by the above formula. And B is the transformation matrix, defined as follows:

Then convert the image from XYZ color space to Lab color space, the conversion formula is as follows:

分别表示的是经过转换后得到的图像中像素点

在Lab颜色空间下每个通道的颜色值。in,

Respectively represent the pixels in the image obtained after conversion

The color value of each channel in the Lab color space.

The purpose of converting the color space is to be more in line with the difference perception of color by human vision, and to facilitate the calculation of salient values.

In step S3, GB segmentation is performed on the segmentation graph obtained in step S1. The core idea is a graph-based greedy clustering algorithm, which concretizes the image as a graph, and performs similar pixels in concepts such as vertices, edges, weights, trees, and minimum spanning trees. The area of the point is divided, so that the image of the segmented area of the block is obtained.

[00] The core of the division rule of pixels is to use the dissimilarity threshold, and its formula is:

Wherein, w _i,j is the largest edge of the current merged region, that is, Int(C _i ∪C _j )=wi _,j . Cx represents each vertex.

In step S4, after obtaining the segmentation map of step S3 and the Lab space map of step S2, a preliminary significant value calculation can be performed according to the area with a short distance and a small influence, and the calculation formula is as follows:

Among them, M(c _k ) represents the significant value of the segmented region c _k ; w(c _i ) is the weight of the region c _i , representing its influence on the significant value of the region c _k ; L _c (c _k ,c _i ) is the spatial distance between the regions _ck and _ci , that is, the added spatial information; σ ² is used to control the strength of the spatial weight, if the value of σ ² is too large, the influence of the spatial weight will also be increases, and the farther area will also have a strong influence on the current area, where the value of σ ² is 0.4; where L _l ( _ck , _ci ) represents the color between the area _ck and _ci Distance metric, and f(n _k,i ) is the frequency of the i-th color in the segmented area _{ck appearing in all colors n k in} this area, k={1,2}; n ₁ and n ₂ represent the area c ₁ respectively and the total number of colors in area c ₂ .

In order to highlight the influence of calculating the wound on the saliency of the image, the wound weight is introduced to calculate the initial saliency value, and the similarity between the wound image and the segmented area in the image set is calculated. The wound weight _w (ci ) is renamed W( _ci ) and its calculation formula is:

Wherein, l( _{ci , c j )} refers to the similarity between the segmented region c _i and the standard wound image c _j .

In the process of calculating the similarity of the wound image, the color feature of the wound and the first-order color distance are used, because this value can reflect the overall sensitivity of the image, and the calculation formula is as follows:

Among them, D represents the number of pixels in a specified area, and x represents the value of pixels in the specified area.

In step S5, the initial saliency map obtained in step S4 is recalculated for the saliency value. Specifically, the moving visual focus is mainly introduced. This is because the center of gravity of most images is not in the center of the image, and the calculation of the saliency value using the image center as the visual focus will have a certain deviation. In order to get the final visual focus, a saliency gravity model is introduced, and each pixel will be affected by gravity, which is divided into horizontal and vertical gravity:

Among them, P refers to the gravitational force, and θ refers to the angle between the current pixel and the visual focus.

Among them, a _x and a _y represent the coordinates of the current pixel (horizontal and vertical directions); VP _x and VP _y represent the coordinates of the visual focus (horizontal and vertical directions); O(a _i , VP) It refers to the Euclidean distance from the current pixel point a _i to the visual focus VP.

To get the final visual focus, it is necessary to calculate the distance between the initial focus and the final visual focus, which is mainly calculated by the movement of the force, and the direction of the two forces is opposite to the direction of the larger force.

Finally, the final visual focus is used to recalculate the saliency value. The final calculation formula is:

J ^f =exp(-τ×L _c (c _k ,VP ^f )) (13)

Among them, L _c (c _k , VP ^f ) represents the spatial distance between the segmented area _ck and the final visual focus VP ^f , τ is used to control the intensity of the visual focus, and the force of increasing the τ visual focus will also increase , resulting in an increase in J ^f , which increases the salient value of the area close to the visual focus, making it easier to extract the area around the visual focus.

Claims (6)

1. A hand wound detection method based on wound weighted significance is characterized by comprising the following steps: wound weighting the saliency images, performing the following steps:

step S1, extracting the foreground of the original image by using GrabCont technology to obtain a segmentation image K for extracting the hand;

step S2, the segmentation chart K obtained in step S1 is initially quantized and subjected to advanced color screening to obtain an image set K with small visual difference and reduced color number after quantization²And performing color space conversion to obtain an image set K³；

Step S3, the segmentation chart K obtained in the step S1 is segmented by using the GB technology to obtain a color segmentation chart C^m；

Step S4, Lab color space image K obtained in step S2³The color segmentation chart C obtained according to the step S3^mCalculating color distance between regions and introducing wound featuresWeighting to obtain an initial saliency map K⁴；

Step S5, the initial saliency map K obtained in step S4⁴Recalculating visual focus of image and actual salient map K of hand wound⁵。

2. The wound-weighted significance-based hand wound detection method of claim 1, wherein: in step S2, the segmentation map K obtained in step S1 is initially quantized, in which the color value of each channel in the image is quantized into 12 different values in RGB color space, and the initial quantization operator is: p is R^m×n→R^m×nFor each pixel of each image, the quantized color value is:

wherein s is 1,2, …, m; t is 1,2, …, n,

and respectively representing the color value of each channel of the pixel points in the original image in the RGB space.

Respectively representing the color channel value of each pixel point in the quantized image in the RGB space; int (.) denotes the conversion of floating point type to integer type.

In order to ensure that the quantized image contains 95% of the pixels of the original image and does not affect the quality, further quantization, i.e. advanced color screening, is required for each pixel

The quantization function is as follows:

wherein,

and respectively representing the color value of each channel of the pixel point in the quantized image in the RGB space.

3. The wound-weighted significance-based hand wound detection method of claim 1, wherein: step S2 of the image set K obtained in step two²The color space conversion is carried out, the RGB color space is converted into the Lab color space, and the main purpose is to better accord with the difference perception of human vision to color and facilitate calculation. The conversion process requires an excess space XYZ.

When converting from RGB color space to XYZ color space, the conversion formula is:

wherein,

respectively represent the pixel points in the image obtained after quantization

Color values of three color channels (RGB) in the RGB color space,

respectively represents the pixel points H in the image obtained by the conversion of the formula_s,tColor values of three color channels (XYZ) in the XYZ color space. And B is a transformation matrix defined as follows:

then, the image is converted from the XYZ color space to the Lab color space, and the conversion formula is as follows:

wherein,

respectively representing the pixel points in the image obtained by conversion

Color values for each channel in Lab color space.

4. The wound-weighted significance-based hand wound detection method of claim 1, wherein: and step S3, performing GB segmentation on the image set K obtained in the step S1, and realizing region judgment on each vertex by the algorithm based on a greedy clustering algorithm of the image to divide the similar pixel points into a region, thereby finally obtaining the segmentation regions with different colors.

5. The wound-weighted significance-based hand wound detection method of claim 1, wherein: the Lab color space image K obtained in step S2³The color segmentation chart C obtained according to the step S3^mCalculating the color distance between the regions, wherein the obvious value calculation formula is as follows:

wherein, M (c)_k) Showing a divided region c_kA significance value of;w(c_i) Is a region c_iThe weight of (c) represents its corresponding region c_kThe effect of significance; l is_c(c_k,c_i) Is a region c_kAnd c_iThe spatial distance between them, i.e. this spatial information added; sigma²For controlling the spatial weight strength, if σ²If the value is too large, the influence strength of the space weight will also increase, and the regions with longer distance will also have stronger influence on the current region, where σ²The value of (A) is 0.4; wherein L is_l(c_k,c_i) Denotes the region c_kAnd c_iThe color distance measure between, and f (n)_k,i) Is a divided region c_kAll colors n in the region of the ith color_kThe frequency of occurrence, k ═ 1,2 }; n is₁And n₂Respectively represent the regions c₁And region c₂Total number of colors of (a).

Calculating the initial significance value introduces wound weight, similarity calculation is carried out on the wound image and the segmentation region in the image set, the more similar weight is larger, the significance of the wound is increased, and the wound weight w (c)_i) Renamed as W (c)_i) And the calculation formula is as follows:

wherein l (c)_i,c_j) Refers to the divided region c_iAnd standard wound image c_jThe similarity of (c).

6. The wound-weighted significance-based hand wound detection method of claim 1, wherein: for the initial saliency map K obtained in step S4⁴Recalculating the visual focus of the image, and obtaining a final hand wound saliency map, wherein the formula is as follows:

J^f＝exp(-τ×L_c(c_k,VP^f)) (9)

wherein L is_c(c_k,VP^f) Showing a divided region c_kAnd the final visual focus VP^fThe spatial distance between τ and τ is used to control the intensity of the visual focus, and the force to increase τ is also greater, resulting in J^fIncreasing thereby increasing the saliency of the area near the visual focus and thus making it easier to focus the area around the visual focus.

Images