Skin texture monitoring method based on image processing

Abstract

The invention discloses a skin texture monitoring method based on image processing, and belongs to the technical field of image processing. The method comprises the following steps: firstly, counting pixel value numbers of skin texture image pixel points in a local area, row and column to obtain a local number, a row number and a column number; according to the local number of the same pixel point and the row and column number difference value, the row and column local difference degree is calculated, and a corresponding difference image is constructed; marking abnormal pixel points from the difference image, and obtaining first and second form attention degrees according to the original image form; performing enhancement processing on the difference image to obtain an enhanced image; and finally, processing the enhanced image by using a dual-channel convolutional neural network, and obtaining the skin texture deviation degree in combination with the form attention degree. According to the method, the local abnormal condition is highlighted by comparing the number of local pixel values with the number of row pixel values and the number of column pixel values, and the accuracy of monitoring results is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a skin texture monitoring method based on image processing. Background Art

[0002] Traditional skin texture monitoring methods mainly rely on manual observation, which not only consumes a large amount of time and energy, but also has strong subjectivity, and there are significant differences in judgments among different observers. With the development of image processing technology, automated skin texture monitoring methods have gradually emerged. Existing skin texture monitoring technologies based on image processing mostly focus on simple texture feature extraction, such as extracting features such as the direction and contrast of texture through a gray-level co-occurrence matrix. However, these methods often only consider the global features of the image and insufficiently mine the features of local regions. In complex skin texture monitoring scenarios, due to the diversity and individual differences of skin textures, global features cannot accurately reflect local detail changes. For example, when monitoring skin aging, lesions, etc., subtle changes in local textures are key information, but traditional methods are difficult to capture these changes in local features, resulting in inaccurate monitoring results and the inability to detect potential skin problems in a timely manner. Summary of the Invention

[0003] Aiming at the above deficiencies in the prior art, a skin texture monitoring method based on image processing provided by the present invention solves the problem of inaccurate monitoring results existing in the prior art.

[0004] To achieve the above invention object, the technical solution adopted by the present invention is: a skin texture monitoring method based on image processing, including the following steps: Count the number of occurrences of the pixel value of each pixel point in the skin texture image in the local area, the row where it is located, and the column where it is located, to obtain the local number, the row number, and the column number; Calculate the row local difference degree and the column local difference degree according to the differences between the local number of the same pixel point and the row number and the column number respectively, and construct a row local difference image and a column local difference image; Mark abnormal pixel points on the row local difference image and the column local difference image, and obtain the first morphological attention degree and the second morphological attention degree based on the morphology of the abnormal pixel points on the skin texture image; Perform difference degree enhancement processing on the row local difference image and the column local difference image respectively to obtain a row local difference enhanced image and a column local difference enhanced image; Process the row local difference enhanced image and the column local difference enhanced image by using a dual-channel convolutional neural network, and obtain the skin texture deviation degree based on the attention degrees imposed by the first morphological attention degree and the second morphological attention degree.

[0005] Further, the specific process of obtaining the local quantity, the number of rows, and the number of columns is as follows: Taking each pixel point in the skin texture image as the center, take a neighborhood, and count the number of occurrences of the pixel value at the central pixel point in the neighborhood to obtain the local quantity. Count the number of occurrences of the pixel value of each pixel point in its corresponding row to obtain the number of rows, and count the number of occurrences of the pixel value of each pixel point in its corresponding column to obtain the number of columns, where N is a positive integer.

[0006] Further, the specific process of constructing the row local difference image and the column local difference image is as follows: Calculate the row local difference degree according to the difference between the local quantity and the number of rows of the same pixel point, and replace the original pixel value with the row local difference degree of the same pixel point to obtain the row local difference image; Calculate the column local difference degree according to the difference between the local quantity and the number of columns of the same pixel point, and replace the original pixel value with the column local difference degree of the same pixel point to obtain the column local difference image.

[0007] Further, the formula for calculating the row local difference degree is: , where is the row local difference degree of the pixel point at the i-th row and the j-th column, is the local quantity of the pixel point at the i-th row and the j-th column, is the number of rows of the pixel point at the i-th row and the j-th column, and e is the natural constant; The formula for calculating the column local difference degree is: , where is the column local difference degree of the pixel point at the i-th row and the j-th column, is the number of columns of the pixel point at the i-th row and the j-th column, and i and j are positive integers.

[0008] Further, the specific process of obtaining the first form attention degree and the second form attention degree is as follows: Calculate the mean value of all row local difference degrees on the row local difference image to obtain the first mean value; On the row local difference image, mark the pixel points with row local difference degrees greater than the first mean value as the first type of abnormal pixel points; Calculate the mean value of all column local difference degrees on the column local difference image to obtain the second mean value; On the column local difference image, mark the pixel points with column local difference degrees greater than the second mean value as the second type of abnormal pixel points; Obtain the first form attention degree according to the positions of the first type of abnormal pixel points on the skin texture image; Obtain the second form attention degree according to the positions of the second type of abnormal pixel points on the skin texture image.

[0009] Further, the specific processes of obtaining the attention degree of the first form and the attention degree of the second form are both as follows: According to the connected regions formed by each abnormal pixel point on the skin texture image, calculate the curvature of each edge pixel point of the connected region; Take the normalized value of the variance of the curvatures of all edge pixel points of the connected region as the edge disorder degree; Calculate the pixel contrast according to the difference between the pixel values on the connected region and the pixel values of other pixel points on the skin texture image; Add the edge disorder degree and the pixel contrast to obtain the morphological feature value of the connected region; Calculate the attention degree of the form according to the area and morphological feature value of each connected region.

[0010] Further, the formula for calculating the pixel contrast is: , where θ n is the pixel contrast of the nth connected region, P n,avg is the average value of the pixel values of the nth connected region, P all,avg is the average value of the pixel values of other pixel points on the skin texture image, | | is the absolute value operation, n is a positive integer, and ε is a constant; The formula for calculating the attention degree of the form is: , where ζ is the attention degree of the form, E n is the area of the nth connected region, γ n is the morphological feature value of the nth connected region, and M is the number of connected regions.

[0011] Further, the formula for enhancing the difference degree of the row local difference image is: , where is the enhanced row local difference degree of the pixel point at the i-th row and the j-th column, is the row local difference degree of the pixel point at the i-th row and the j-th column, D R,max is the maximum row local difference degree; The formula for enhancing the difference degree of the column local difference image is: , where is the enhanced column local difference degree of the pixel point at the i-th row and the j-th column, is the column local difference degree of the pixel point at the i-th row and the j-th column, D C,max is the maximum column local difference degree.

[0012] Further, the dual-channel convolutional neural network includes: a first image feature extraction channel, a second image feature extraction channel, a first attention application layer, a second attention application layer, a first convolutional block, a second convolutional block, a feature fusion layer, and a fully connected layer; The input end of the first image feature extraction channel is used to input the row local difference enhanced image; The input end of the second image feature extraction channel is used to input the column local difference enhanced image; The first input end of the first attention application layer is connected to the output end of the first image feature extraction channel, its second input end is used to input the first morphological attention, and its output end is connected to the input end of the first convolutional block; The first input end of the second attention application layer is connected to the output end of the second image feature extraction channel, its second input end is used to input the second morphological attention, and its output end is connected to the input end of the second convolutional block; The input ends of the feature fusion layer are respectively connected to the output ends of the first convolutional block and the second convolutional block, and its output end is connected to the input end of the fully connected layer; The output end of the fully connected layer serves as the output end of the dual-channel convolutional neural network.

[0013] Furthermore, the expression of the first attention application layer is: , where G 1 is the output of the first attention application layer, ζ 1 is the first morphological attention, and S 1 is the output of the first image feature extraction channel; The expression of the second attention application layer is: , where G 2 is the output of the second attention application layer, ζ 2 is the second morphological attention, and S 2 is the output of the second image feature extraction channel.

[0014] The beneficial effects of the present invention are as follows: 1. By counting the number of occurrences of each pixel point in the local area, the row where it is located, and the column where it is located, the present invention can carefully analyze the local characteristics of the image. The present invention considers pixel points from multiple dimensions (local, row, column), captures local abnormalities in skin texture, and improves the sensitivity to local texture changes.

[0015] 2. The present invention constructs a difference image based on the difference between the local quantity of the same pixel point and the row and column quantities. This method highlights the differences between the local area and the row quantity as well as the column quantity, thereby measuring the uniformity of texture distribution and highlighting abnormal areas.

[0016] 3. The present invention marks abnormal pixel points on the difference image and obtains attention according to their morphology on the original skin texture image, which helps to more accurately identify real abnormalities.

[0017] 4. The present invention performs a difference degree enhancement process on the row and column local difference images, further highlighting the difference between abnormal textures and normal textures.

[0018] 5. The present invention uses a dual-channel convolutional neural network to process the enhanced difference image, which can make full use of the spatial information and features of the image. Combining the attentions imposed by the first morphological attention degree and the second morphological attention degree, it can calculate the skin texture deviation degree more accurately, so as to comprehensively and accurately evaluate the skin texture state and improve the accuracy of skin monitoring. Description of the Drawings

[0019] Figure 1 is a flowchart of a skin texture monitoring method based on image processing; Figure 2 is an example schematic diagram of a neighborhood; Figure 3 is a schematic structural diagram of a dual-channel convolutional neural network. Detailed Embodiments

[0020] The following describes the detailed embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the detailed embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions made using the concept of the present invention are within the scope of protection.

[0021] As Figure 1 shown, a skin texture monitoring method based on image processing includes the following steps: Count the number of occurrences of the pixel value of each pixel point in the skin texture image in the local area, the row where it is located, and the column where it is located to obtain the local number, the row number, and the column number; According to the differences between the local number of the same pixel point and the row number and the column number respectively, calculate the row local difference degree and the column local difference degree, and construct a row local difference image and a column local difference image; Mark abnormal pixel points on the row local difference image and the column local difference image, and based on the morphology of the abnormal pixel points on the skin texture image, obtain the first morphological attention degree and the second morphological attention degree; Perform difference degree enhancement processing on the row local difference image and the column local difference image respectively to obtain a row local difference enhanced image and a column local difference enhanced image; Use a dual-channel convolutional neural network to process the row local difference enhanced image and the column local difference enhanced image, and based on the attentions imposed by the first morphological attention degree and the second morphological attention degree, obtain the skin texture deviation degree.

[0022] In this embodiment, the specific process of obtaining the local number, the row number, and the column number is as follows: Taking each pixel point in the skin texture image as the center, take a neighborhood, and count the pixel value at the center pixel point in The number of occurrences in the neighborhood is obtained to obtain the local number, the number of occurrences of the pixel value of each pixel point in the row is counted to obtain the row number, and the number of occurrences of the pixel value of each pixel point in the column is counted to obtain the column number, where N is a positive integer.

[0023] For example, taking the pixel at the 5th row and 3rd column as the center, the pixel value at the center pixel is The number of occurrences in the neighborhood is used to obtain the local number, and then the number of pixel values ​​at the center pixel point in the 5th row is counted, and then the number of pixel values ​​at the center pixel point in the 3rd column is counted.

[0024] In this embodiment, the skin texture image is Resolution, 1920 pixels per row and 1080 pixels per column.

[0025] In this embodiment, the skin texture image may be an image captured by a dermatoscope and is not limited to the dermatoscope image mentioned in this embodiment, and any other image that can reflect the skin texture state may be used.

[0026] In this embodiment, Set to or Wait, set it to When The distribution difference between the pixel values ​​in the range and the pixel values ​​in a row or column is set to When The distribution difference between the pixel values ​​in the range and the pixel values ​​in a row or column. N is set smaller, which can better compare the pixel values ​​of the local area with those of a row or column, and can better explore the characteristics of the local area. The size of N set in the present invention is not limited, and the size of N can be adjusted according to different skin detection conditions. In the specific operation process, different N can be set to obtain the skin texture deviation under different N.

[0027] For example, Neighborhood Figure 2 As shown, the pixel value of the center pixel is 120, then In the neighborhood, 120 appears 7 times. Check the row where the pixel at the center is located (row 100). There are 1920 pixels in this row. Select some pixel values ​​to display: 116, 119, 120, 122, 124, 125, 115, …, 118, 120, 120, 123, 125, 127, 130. Count the number of times 120 appears. When counting the local number, row number, and column number, the pixel value of the pixel at the center is included. The pixel value in the example uses the image grayscale value.

[0028] Normal skin texture images are consistent both locally and globally, while abnormal texture changes lead to an increased difference between the local and the whole. By calculating the row local difference degree and the column local difference degree, the areas with abnormal texture can be highlighted.

[0029] In this embodiment, the specific process of constructing the row local difference image and the column local difference image is as follows: According to the difference between the local quantity of the same pixel point and the number of rows, calculate the row local difference degree, and replace the original pixel value with the row local difference degree of the same pixel point to obtain the row local difference image; According to the difference between the local quantity of the same pixel point and the number of columns, calculate the column local difference degree, and replace the original pixel value with the column local difference degree of the same pixel point to obtain the column local difference image.

[0030] In this embodiment, the difference between the local quantity of the same pixel point and the number of rows can be used as the row local difference degree, and the difference between the local quantity of the same pixel point and the number of columns can be used as the column local difference degree. More preferably, the formula for calculating the row local difference degree is: , where is the row local difference degree of the pixel point at the i-th row and j-th column, is the local quantity of the pixel point at the i-th row and j-th column, is the number of rows of the pixel point at the i-th row and j-th column, and e is the natural constant; The formula for calculating the column local difference degree is: , where is the column local difference degree of the pixel point at the i-th row and j-th column, is the number of columns of the pixel point at the i-th row and j-th column, and i and j are positive integers.

[0031] By calculating the differences between the local quantity of the same pixel point and the number of rows and columns, and constructing the difference degree formula in exponential form, the present invention can magnify the difference between the local and global distributions of pixel points. Normal skin texture has a certain regularity and consistency, while when there are lesions, aging, etc., the local texture will deviate from the overall pattern. This calculation method can sensitively capture these deviations and significantly reflect the differences.

[0032] At greater than , greater than , the values of the row local difference degree and the column local difference degree are larger. While at less than , less than , the values of the row local difference degree and the column local difference degree are between 0 and 1. Therefore, the abnormal and normal values can be significantly distinguished, facilitating the acquisition of the first form attention degree and the second form attention degree.

[0033] In this embodiment, the specific process of obtaining the first form attention degree and the second form attention degree is as follows: Calculate the mean value of all row local difference degrees on the row local difference image to obtain the first mean value; On the row local difference image, mark the pixel points with row local difference degrees greater than the first mean value as the first type of abnormal pixel points; Calculate the mean value of all column local difference degrees on the column local difference image to obtain the second mean value; On the column local difference image, mark the pixel points with column local difference degrees greater than the second mean value as the second type of abnormal pixel points; Obtain the first form attention degree according to the positions of the first type of abnormal pixel points on the skin texture image; Obtain the second form attention degree according to the positions of the second type of abnormal pixel points on the skin texture image.

[0034] The present invention determines the threshold by calculating the mean values of the row and column local difference images, marks the pixel points greater than the mean value as abnormal pixel points, and applies statistical principles. In image processing, the mean value can reflect the central tendency of data, and the pixel points greater than the mean value can be regarded as abnormal points deviating from the normal distribution. This screening method accurately locates the pixel points with texture abnormalities at the data level.

[0035] The abnormalities of skin texture are not evenly distributed but concentrated in certain specific areas. The abnormal pixel points screened by the above method can better capture these abnormal areas. For example, when wrinkles or skin spots appear on the skin, the texture in the corresponding area will show differences in the local and overall comparison, and will be marked as abnormal pixel points. Furthermore, the obtained form attention degree can reflect the distribution form of these abnormal areas.

[0036] In this embodiment, the specific process of obtaining the first form attention degree and the second form attention degree is as follows: According to the connected regions formed by each abnormal pixel point on the skin texture image, calculate the curvature of each edge pixel point of the connected region; Take the normalized value of the variance of the curvatures of all edge pixel points of the connected region as the edge clutter degree; Calculate the pixel contrast according to the difference between the pixel values on the connected region and the pixel values of other pixel points on the skin texture image; Add the edge clutter degree and the pixel contrast to obtain the form feature value of the connected region; Calculate the form attention degree according to the areas and form feature values of each connected region.

[0037] In the present invention, the connected region is that each abnormal pixel point is adjacent and connected in space position to form a continuous region.

[0038] The curvature reflects the degree of bending of the edge, and the variance normalization value reflects the change of the overall edge. In image processing, this method can accurately capture the contour features of abnormal areas of skin texture. At the same time, calculating the pixel contrast measures the difference between the abnormal area and the surrounding area from the aspect of pixel difference. The combination of the two comprehensively describes the morphological features of the connected area.

[0039] In this embodiment, the formula for calculating the pixel contrast is: , where θ n is the pixel contrast of the nth connected area, P n,avg is the mean value of the pixel values of the nth connected area, P all,avg is the mean value of the pixel values of other pixel points on the skin texture image, | | is the absolute value operation, n is a positive integer, and ε is a constant.

[0040] In the formula for calculating the pixel contrast of the present invention, other pixel points on the skin texture image refer to pixel points except for each abnormal pixel point.

[0041] The constant ε is used to avoid the denominator being 0.

[0042] The present invention obtains the pixel contrast by finding the difference between the mean value of the pixels in the connected area and the mean value of other pixels, and normalizing the difference, so as to measure the difference degree between the connected area and other parts of the skin texture image.

[0043] In this embodiment, the mean value of the morphological feature values of each connected area can be used as the morphological attention degree. More preferably, the formula for calculating the morphological attention degree is: , where ζ is the morphological attention degree, E n is the area of the nth connected area, γ n is the morphological feature value of the nth connected area, and M is the number of connected areas.

[0044] The present invention uses the area of each connected area as a weight to calculate the centroid of each connected area, which is convenient for measuring the overall morphological situation. In this way, the connected area with a larger area proportion can obtain a larger proportion. Compared with the mean value, the centroid can better reflect the concentration trend of the abnormal distribution of skin texture and highlight the dominant role of the large-area abnormal area in the overall morphology.

[0045] In this embodiment, the formula for enhancing the difference degree of the row local difference image is: , where, is the enhanced row local difference degree of the pixel point at the i-th row and j-th column, is the row local difference degree of the pixel point at the i-th row and j-th column, D R,max is the maximum row local difference degree; The formula for enhancing the degree of difference of the column local difference image is as follows: , where is the enhanced column local difference degree of the pixel at the i-th row and j-th column, is the column local difference degree of the pixel at the i-th row and j-th column, and D C,max is the maximum column local difference degree.

[0046] In the present invention, by performing a square operation on the row and column local difference degrees and utilizing the characteristics of the square function, the original difference degree value can be effectively amplified. For the pixel points with a relatively large original difference degree (corresponding to the abnormal skin texture area), the enhanced difference degree will increase significantly; while for the pixel points with a small difference degree (normal texture area), the change is relatively small. In this way, the difference degree gap between the normal and abnormal areas is further widened, making the abnormal texture part more prominent in the image and facilitating the subsequent accurate identification of texture abnormalities. For example, in a local area, the occurrence number of the pixel value at the center point is 5, while the occurrence number of this pixel value in the same row is only 1, then the row local difference degree is e 4 ; in a local area, the occurrence number of the pixel value at the center point is 5, while the occurrence number of this pixel value in the same row is 100, then the row local difference degree is e -95 , and then by squaring e 4 and e -95 , the gap between the two is further widened. Normalization is to reduce the amount of data processed by the dual-channel convolutional neural network and at the same time make each value on the same scale.

[0047] As Figure 3 shown, the dual-channel convolutional neural network includes: a first image feature extraction channel, a second image feature extraction channel, a first attention application layer, a second attention application layer, a first convolutional block, a second convolutional block, a feature fusion layer, and a fully connected layer; The input end of the first image feature extraction channel is used to input the row local difference enhanced image; The input end of the second image feature extraction channel is used to input the column local difference enhanced image; The first input end of the first attention application layer is connected to the output end of the first image feature extraction channel, its second input end is used to input the first morphological attention, and its output end is connected to the input end of the first convolutional block; the first input end of the second attention application layer is connected to the output end of the second image feature extraction channel, its second input end is used to input the second morphological attention, and its output end is connected to the input end of the second convolutional block; the input ends of the feature fusion layer are respectively connected to the output ends of the first convolutional block and the second convolutional block, and its output end is connected to the input end of the fully connected layer; The output end of the fully connected layer serves as the output end of the dual-channel convolutional neural network.

[0048] In this embodiment, both of the two image feature extraction channels are: CNN neural networks.

[0049] In this embodiment, the feature fusion layer adds the outputs of the first convolutional block and the second convolutional block element-wise to obtain the fused features.

[0050] The expression of the first attention application layer is: , where G 1 is the output of the first attention application layer, ζ 1 is the first morphological attention, S 1 is the output of the first image feature extraction channel; The expression of the second attention application layer is: , where G 2 is the output of the second attention application layer, ζ 2 is the second morphological attention, S 2 is the output of the second image feature extraction channel.

[0051] The present invention uses two independent image feature extraction channels to process the row and column local difference images respectively, which can capture the texture difference features from two dimensions, avoid information omission caused by a single channel, and improve the integrity and accuracy of feature extraction.

[0052] The present invention combines the first and second morphological attentions through the first and second attention application layers, enabling the network to enhance the attention to important features when processing images. In the present invention, the larger the first and second morphological attentions are, the greater the possibility of the existence of abnormal regions in the skin texture image. Multiplying the morphological attention by the output of the image feature extraction channel focuses more on the important texture difference features and suppresses the interference of irrelevant or secondary features, which is beneficial to improving the recognition accuracy of the network for texture abnormalities.

[0053] The present invention can carefully analyze the local characteristics of the image by counting the number of occurrences of each pixel point in the local area, the row where it is located, and the column where it is located. The present invention considers pixel points from multiple dimensions (local, row, column), captures the local abnormalities of the skin texture, and improves the sensitivity to local texture changes.

[0054] The present invention constructs a difference image based on the difference between the local quantity and the row and column quantities of the same pixel point. This method highlights the differences between the local area and the row quantity as well as the column quantity, thereby measuring the uniformity of the texture distribution and highlighting the abnormal regions.

[0055] The present invention marks abnormal pixel points on the difference image and obtains the attention according to their morphology on the original skin texture image, which helps to more accurately identify the true abnormalities.

[0056] The present invention performs a difference degree enhancement process on the row and column local difference images, further highlighting the difference between abnormal textures and normal textures.

[0057] The present invention uses a dual-channel convolutional neural network to process the enhanced difference image, which can make full use of the spatial information and features of the image, and combines the attention degrees imposed by the first morphological attention degree and the second morphological attention degree to more accurately calculate the skin texture deviation degree, so as to comprehensively and accurately evaluate the skin texture state.

[0058] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A skin texture monitoring method based on image processing, characterized in that: The following steps are involved: Count the number of occurrences of the pixel value of each pixel point in the skin texture image in the local area, the row, and the column to obtain the local number, the row number, and the column number; According to the difference between the local number of the same pixel and the number of rows and the number of columns, the row local difference and the column local difference are calculated, and the row local difference image and the column local difference image are constructed; Mark abnormal pixel points on the row local difference image and the column local difference image, and obtain a first form attention degree and a second form attention degree based on the form of the abnormal pixel points on the skin texture image; Performing difference enhancement processing on the row local difference image and the column local difference image respectively to obtain a row local difference enhanced image and a column local difference enhanced image; A dual-channel convolutional neural network is used to process the row local difference enhanced image and the column local difference enhanced image, and the skin texture deviation is obtained based on the attention imposed by the first form of attention and the second form of attention.

2. The skin texture monitoring method based on image processing according to claim 1, characterized in that: The specific process of obtaining the local number, row number, and column number is as follows: take each pixel point in the skin texture image as the center and Neighborhood, the pixel value at the statistical center pixel is The number of occurrences in the neighborhood is obtained to obtain the local number, the number of occurrences of the pixel value of each pixel point in the row is counted to obtain the row number, and the number of occurrences of the pixel value of each pixel point in the column is counted to obtain the column number, where N is a positive integer.

3. The skin texture monitoring method based on image processing according to claim 1, characterized in that: The specific process of constructing the row local difference image and the column local difference image is as follows: according to the difference between the local number and the row number of the same pixel point, the row local difference degree is calculated, and the row local difference degree of the same pixel point is used to replace the original pixel value to obtain the row local difference image; According to the difference between the local number and the column number of the same pixel, the column local difference is calculated, and the column local difference of the same pixel is used to replace the original pixel value to obtain the column local difference image.

4. The skin texture monitoring method based on image processing according to claim 1, characterized in that: The formula for calculating the local difference of a row is: ,in, is the row local difference of the pixel point in the i-th row and j-th column, is the local number of pixels in the i-th row and j-th column, is the number of rows of pixels in the i-th row and j-th column, and e is a natural constant; The formula for calculating the local difference of a column is: ,in, is the column local difference of the pixel point in the i-th row and j-th column, is the column number of the pixel in the i-th row and j-th column, where i and j are positive integers.

5. The skin texture monitoring method based on image processing according to claim 1, characterized in that: The specific process of obtaining the first form of attention and the second form of attention is: Calculating the mean of all row local differences on the row local difference image to obtain a first mean; Marking pixels whose row local difference is greater than the first mean value as first-category abnormal pixels on the row local difference image; Calculating the mean of all column local differences on the column local difference image to obtain a second mean; On the column local difference image, pixels whose column local difference is greater than the second mean are marked as second-category abnormal pixels; According to the position of the first type of abnormal pixel points on the skin texture image, a first form attention degree is obtained; According to the positions of the second type of abnormal pixel points on the skin texture image, the second form attention degree is obtained.

6. The skin texture monitoring method based on image processing according to claim 5, characterized in that: The specific processes of obtaining the first form of attention and the second form of attention are: According to the connected area formed by each abnormal pixel point on the skin texture image, the curvature of each edge pixel point in the connected area is calculated; The normalized value of the variance of the curvature of all edge pixels in the connected area is taken as the edge clutter; Calculate pixel contrast based on the difference between the pixel value in the connected area and the pixel values ​​of other pixels in the skin texture image; The edge clutter and pixel contrast are added to obtain the morphological feature value of the connected area; The morphological attention is calculated based on the area and morphological feature values ​​of each connected region.

7. The skin texture monitoring method based on image processing according to claim 6, characterized in that: The formula for calculating pixel contrast is: , where θ n is the pixel contrast of the nth connected region, P n,avg is the mean pixel value of the nth connected region, P all,avg is the mean value of the pixel values ​​of other pixels on the skin texture image, | | is the absolute value operation, n is a positive integer, and ε is a constant; The formula for calculating morphological attention is: , where ζ is the morphological attention, E n is the area of ​​the nth connected region, γ n is the morphological feature value of the nth connected region, and M is the number of connected regions.

8. The skin texture monitoring method based on image processing according to claim 1, characterized in that: The formula for enhancing the difference of the row local difference image is: ,in, is the enhanced row local difference of the pixel point in the i-th row and j-th column, is the row local difference of the pixel point in the i-th row and j-th column, D R,max is the maximum row local difference, i and j are positive integers; The formula for enhancing the difference of the column local difference image is: ,in, is the enhanced column local difference of the pixel point in the i-th row and j-th column, is the column local difference of the pixel point in the i-th row and j-th column, D C,max is the maximum column local difference.

9. The skin texture monitoring method based on image processing according to claim 1, characterized in that: The dual-channel convolutional neural network includes: a first image feature extraction channel, a second image feature extraction channel, a first attention application layer, a second attention application layer, a first convolution block, a second convolution block, a feature fusion layer and a fully connected layer; The input end of the first image feature extraction channel is used to input the row local difference enhanced image; The input end of the second image feature extraction channel is used to input the column local difference enhanced image; The first input end of the first attention applying layer is connected to the output end of the first image feature extraction channel, the second input end thereof is used to input the first morphological attention, and the output end thereof is connected to the input end of the first convolution block; A first input end of the second attention applying layer is connected to an output end of the second image feature extraction channel, a second input end thereof is used to input a second form of attention, and an output end thereof is connected to an input end of the second convolution block; The input end of the feature fusion layer is connected to the output end of the first convolution block and the output end of the second convolution block respectively, and its output end is connected to the input end of the fully connected layer; The output of the fully connected layer is used as the output of the two-channel convolutional neural network.

10. The skin texture monitoring method based on image processing according to claim 9, characterized in that: The expression of the first attention application layer is: , where G1 is the output of the first attention application layer, ζ1 is the first morphological attention, and S1 is the output of the first image feature extraction channel; The expression of the second attention application layer is: , where G2 is the output of the second attention application layer, ζ2 is the second form of attention, and S2 is the output of the second image feature extraction channel.

Images