A method and system for identifying local areas in intense pulsed light skin beauty

Abstract

The present invention provides a method and system for identifying local areas of intense pulsed light skin beauty, which relates to the field of phototherapy technology. The above solution includes collecting the original facial image of a patient and obtaining a historical skin treatment segmentation data set, preprocessing the original facial image to obtain a standard facial image; constructing a skin area division model based on the image feature information in the skin treatment segmentation data set, inputting the standard facial image into the skin area division model for target recognition to obtain multiple images of areas to be treated in the standard facial image; performing independent component analysis on the images of areas to be treated to obtain the melanin channel and hemoglobin channel in the images of areas to be treated; constructing a multi-modal attention network model, and inputting the melanin channel and hemoglobin channel into the multi-modal attention network model for feature extraction and segmentation to respectively obtain melanin segmentation information and hemoglobin segmentation information; fusing the melanin segmentation information and hemoglobin segmentation information to obtain a skin treatment segmentation map, collecting the treatment features in the skin treatment segmentation map, classifying the treatment features according to a random forest classifier to obtain the skin treatment type corresponding to each image of an area to be treated, and giving a treatment plan corresponding to the skin treatment type. The present invention helps to improve the local detail sensitivity and global information processing efficiency of the skin area division model.

Description

Technical Field

[0001] The present invention relates to the technical field of phototherapy, and particularly to a method and system for identifying local areas of intense pulsed light skin beauty. Background Art

[0002] Intense pulsed light (IPL), also known as pulsed intense light, is a high-intensity light emitted by a special light source. After being focused and filtered, it forms light within a specific wavelength range. This light is non-coherent ordinary light, different from lasers, and its wavelength is usually between 500 and 1200 nm. In the field of epidermal treatment, intense pulsed light is widely used mainly because it can effectively treat a variety of skin problems, especially skin diseases related to phototherapy and photoaging. Intense pulsed light is mainly applied to the treatment of pigmentary skin diseases, vascular skin diseases, and diseases related to changes in dermal collagen tissue structure.

[0003] Chinese Patent with Publication No. CN115100107B discloses a method and system for dermoscopic image segmentation. The method includes obtaining a dermoscopic image and preprocessing the dermoscopic image; based on the preprocessed dermoscopic image, using a segmentation model to obtain a segmentation result. The segmentation model includes: using a ResNet network to extract a first feature map; taking the first feature map as the input of a Transformer, adding a boundary gate in the Transformer to extract local details to handle fuzzy boundaries to obtain a second feature map; based on the second feature map, using a DenseASPP network to enhance feature representation and process multi-scale information to obtain a third feature map, and upsampling the obtained result to restore the resolution to obtain the segmentation result. However, the solution provided in the above application can only generate a limited receptive field and cannot capture the dependencies between global relationships, thus resulting in the inability to effectively extract sufficient local details to handle fuzzy boundaries. Therefore, it is very necessary to provide a method and system for identifying local areas of intense pulsed light skin beauty to improve the local detail sensitivity and global information processing efficiency of the skin area division model. Summary of the Invention

[0004] In view of this, the present invention proposes a method and system for identifying local areas of intense pulsed light skin beauty. By combining the advantages of a skin area division model and independent component analysis, the local detail sensitivity and global information processing efficiency of the skin area division model are improved.

[0005] The present invention provides a method for identifying local areas of intense pulsed light skin beauty, the method comprising:

[0006] Collecting the original facial image of a patient and obtaining a historical skin treatment segmentation data set, preprocessing the original facial image to obtain a standard facial image;

[0007] Based on the image feature information in the skin treatment segmentation dataset, construct a skin area division model, input the facial standard image into the skin area division model for target recognition, and obtain multiple images of areas to be treated in the facial standard image;

[0008] Perform independent component analysis on the images of areas to be treated to obtain the melanin channel and hemoglobin channel in the images of areas to be treated;

[0009] Construct a multi-modal attention network model, and input the melanin channel and the hemoglobin channel into the multi-modal attention network model for feature extraction and segmentation, respectively obtaining melanin segmentation information and hemoglobin segmentation information;

[0010] Fuse the melanin segmentation information and the hemoglobin segmentation information to obtain a skin treatment segmentation map, collect the repair features in the skin treatment segmentation map, classify the repair features according to a random forest classifier, obtain the skin treatment type corresponding to each image of the area to be treated, and give a treatment plan corresponding to the skin treatment type.

[0011] Based on the above technical solutions, preferably, the preprocessing of the original facial image to obtain a facial standard image specifically includes:

[0012] Convert the original facial image to the RBG format, and adjust the facial area in the original facial image to a horizontal state to obtain a facial transition image;

[0013] Perform color correction, noise removal, illumination equalization, and detail enhancement processing on the facial transition image in sequence to obtain a facial standard image.

[0014] Based on the above technical solutions, preferably, the detail enhancement specifically includes:

[0015] Perform segmentation processing on the facial transition image after color correction, noise removal, and illumination equalization processing to obtain a facial transition segmentation map composed of multiple segmentation blocks;

[0016] Obtain the single-channel grayscale image of each segmentation block in the facial transition segmentation map, according to the grayscale level standard, obtain the grayscale level of each pixel point in each segmentation block, and use the pixel points with grayscale levels lower than the preset level threshold as under-enhanced pixel points;

[0017] Adjust the pixel value of the pixel point corresponding to the under-enhanced pixel point in the facial transition image to the corresponding expected pixel value to obtain the facial standard image, where the expected pixel value is obtained based on the grayscale level of the corresponding pixel point, the total number of pixel values of the segmentation block where the corresponding pixel point is located, and the total number of grayscale levels.

[0018] More preferably, constructing a skin region division model based on the image feature information in the skin treatment segmentation dataset specifically includes:

[0019] Obtain the unimodal image data and multimodal image data in the skin treatment segmentation dataset, divide the unimodal image data and multimodal image data into first image data and second image data according to a preset ratio, and extract the data features in the first image data and the second image data based on a feature extraction function;

[0020] Based on the data features in the first image data, construct an initial multimodal attention fusion network, and use multi-scale feature fusion to increase the receptive field of the initial multimodal attention fusion network;

[0021] Extract the image depth features of the initial multimodal attention fusion network, and use the attention mechanism to calibrate the image depth feature channels in the initial multimodal attention fusion network to obtain the calibrated image depth features;

[0022] Fuse the calibrated image depth features with the original facial image to construct a multimodal attention fusion network, and train the multimodal attention fusion network according to the data features and loss function in the second image data to obtain the skin region division model.

[0023] More preferably, the expressions of the feature extraction function and the loss function are respectively:

[0024] ;

[0025] ;

[0026] Among them, M c ( F ) represents the feature extraction function, F represents the feature map input to the attention mechanism module, σ ( ) represents the activation function, MLP ( ) represents the optimization function built in the three-layer perceptron, AvgPool ( ) represents the average pooling function, W 0 represents the first weight matrix, W 1 represents the second weight matrix, represents the feature after average pooling, L total represents the loss function, α represents the weight loss parameter, q represents the true label value output by the model, pRepresents the predicted value of the model output, δ Represents the edge adjustment parameter, N Represents the total number of samples input into the model, n Represents the n th sample, ω Represents the hyperparameter, q n Represents the n true label value of the th sample, p n Represents the n predicted value of the th sample.

[0027] More preferably, performing independent component analysis on the image of the area to be treated to obtain the melanin channel and hemoglobin channel in the image of the area to be treated specifically includes:

[0028] Constructing a skin optical model based on the skin color density function, and introducing the skin optical model and the predicted pure color density vector into the skin optical model to obtain the synthetic skin color density function;

[0029] Based on the synthetic skin color density function and the least squares method, respectively solving the melanin channel and hemoglobin channel in the image of the area to be treated.

[0030] More preferably, the expression of the synthetic skin color density function is:

[0031] ;

[0032] ;

[0033] Wherein, I x,y Represents the skin color density function, x Represents the abscissa of the pixel point, y Represents the abscissa of the pixel point, c m Represents the pure color feature of melanin per unit density, c h Represents the pure color feature of hemoglobin per unit density, Represents the relative amount of melanin in each pixel point, Represents the relative amount of hemoglobin in each pixel point, Δ h Represents the stationary column vector generated by other skin pigments and structures, S x,y Represents the synthetic skin color density function, Represents the predicted pure color density vector, K Represents the first density control parameter, j Represents the second density control parameter.

[0034] More preferably, the random forest classifier includes a feature integration unit, a data set division unit, a decision tree construction unit, a multi-decision tree training unit, and a classification prediction unit. Among them,

[0035] The feature integration unit is used to integrate the repair features collected in the skin treatment segmentation map and form a repair feature data set;

[0036] The data set division unit is used to divide the repair feature data set into several different repair feature sub-data sets by random sampling;

[0037] The decision tree construction unit is used to generate a decision tree and perform recursive splitting;

[0038] The multi-decision tree training unit is used to establish different decision trees on different repair feature sub-data sets respectively;

[0039] The classification prediction unit is used to perform sample classification and output the skin treatment type corresponding to each image of the area to be treated.

[0040] More preferably, the skin treatment types include pigmentary diseases, vascular diseases, keratosis pilaris, acne and its sequelae, and skin texture problems.

[0041] In the second aspect of the present application, a local area recognition system for intense pulsed light skin beauty is provided. The local area recognition system includes a data acquisition module, an information extraction module, and a treatment operation module. Among them,

[0042] The data acquisition module is used to collect the original facial image of the patient and obtain the historical skin treatment segmentation data set, preprocess the original facial image, and obtain the standard facial image;

[0043] The information extraction module is used to construct a skin area division model based on the image feature information in the skin treatment segmentation data set, input the standard facial image into the skin area division model for target recognition, obtain multiple images of the areas to be treated in the standard facial image, perform independent component analysis on the images of the areas to be treated, obtain the melanin channel and hemoglobin channel in the images of the areas to be treated, construct a multi-modal attention network model, and input the melanin channel and the hemoglobin channel into the multi-modal attention network model for feature extraction and segmentation, respectively obtaining melanin segmentation information and hemoglobin segmentation information;

[0044] The treatment operation module is used to fuse the melanin segmentation information and the hemoglobin segmentation information to obtain a skin treatment segmentation map, collect the repair features in the skin treatment segmentation map, classify the repair features according to a random forest classifier, obtain the skin treatment type corresponding to each image of the area to be treated, and give a treatment plan corresponding to the skin treatment type.

[0045] The method and system for local area recognition of intense pulsed light skin beauty provided by the present invention have the following beneficial effects compared with the prior art:

[0046] (1) By collecting the original facial image and combining it with the historical treatment data set to construct a skin area division model, multiple areas to be treated on the face can be accurately recognized and divided. At the same time, independent component analysis is performed on the areas to be treated to separate the melanin channel and the hemoglobin channel, which can analyze the skin pigmentation and vascular characteristics more deeply. The multi-modal attention network model is used to extract and segment the features of the melanin and hemoglobin channels, which can consider both pigment and vascular features at the same time. This multi-modal analysis can provide more comprehensive diagnostic information, thereby improving the accuracy of segmentation. And it can further optimize the features extracted by independent component analysis by taking advantage of the long-term dependence modeling of the skin area division model to complete the task of image segmentation of the areas to be treated. Combining the advantages of the skin area division model and independent component analysis helps to improve the local detail sensitivity and global information processing efficiency of the skin area division model, so as to understand the image content more comprehensively;

[0047] (2) By using the feature extraction function to process different types of image data, more comprehensive skin feature information can be captured. And by multi-scale feature fusion, the receptive field of the network is increased, enabling the model to pay attention to local details and global structures at the same time. The attention mechanism is used to calibrate the image depth feature channels, highlighting important features and suppressing irrelevant information to improve the recognition ability of the skin area division model for key skin features, thereby improving the accuracy of division;

[0048] (3) By constructing a skin optical model, the distributions of melanin and hemoglobin can be separated more accurately. And by introducing the skin optical model and the predicted pure color density vector, the complexity and multi-level nature of the skin structure are considered, making the analysis result closer to the real skin condition. At the same time, it can accurately separate the melanin and hemoglobin channels and accurately quantify the distributions of melanin and hemoglobin, and can more precisely select the parameters of intense pulsed light treatment. Description of the Drawings

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0050] Figure 1 It is a schematic flowchart of a method for identifying local areas in intense pulsed light skin beauty provided by the present invention;

[0051] Figure 2 It is a schematic structural diagram of a local area identification system provided by the present invention.

[0052] Explanation of reference numerals: 1. Local area identification system; 11. Data acquisition module; 12. Information extraction module; 13. Treatment operation module. Detailed implementation manners

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in combination with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0054] The embodiments of the present application disclose a method for identifying local areas in intense pulsed light skin beauty. As Figure 1 shown, the steps of this method include S1 to S5.

[0055] Step S1, collect the original facial image of the patient and obtain the historical skin treatment segmentation data set, and preprocess the original facial image to obtain the standard facial image.

[0056] In this embodiment, a high-resolution digital camera or a professional medical imaging device can be used to collect the original facial image of the patient. For example, a Canon EOS R5 full-frame mirrorless camera equipped with a 45MP high-resolution sensor or a VISIA skin analyzer can be used to take pictures under standard lighting conditions to ensure consistent image quality, and multiple angles are taken to obtain comprehensive information.

[0057] In this step, it includes steps S11 to S12.

[0058] Step S11, convert the original facial image to the RBG format and adjust the facial area in the original facial image to a horizontal state to obtain the facial transition image.

[0059] In this step, use an image processing library to read the original facial image, check the current format of the original facial image. If it is already in RGB format, skip the conversion step; if it is not in RGB format, call the corresponding color space conversion function to convert the original facial image to RGB format and save the converted RGB format image. Apply a face detection algorithm to the RGB format image to locate the facial area. Within the detected facial area, use a feature point detection algorithm to locate the positions of the two eyes. According to the coordinates of the left and right eyes, calculate the facial tilt angle. Use the center of the RGB format image as the rotation center and rotate the entire RGB format image according to the calculated angle. After rotation, re-detect the facial area to ensure that the face is adjusted to a horizontal state. If necessary, crop the image to remove the blank area generated by rotation, and save the adjusted image as the facial transition image.

[0060] Step S12: Perform color correction, noise removal, illumination equalization, and detail enhancement on the facial transition image in sequence to obtain the standard facial image.

[0061] In this step, use the gray world algorithm to calculate the average values of the R, G, and B channels in the image. Assume the average value of R is 120, G is 140, and B is 160. Calculate the adjustment coefficient K R = 140 / 120 = 1.17, K G = 1, K B = 140 / 160 = 0.875. Multiply the R value of each pixel in the image by 1.17, multiply the B value by 0.875, and keep the G value unchanged. Convert the image to the HSV color space, analyze the average value of the S channel (saturation), assume it is 0.4 (range 0 - 1), increase the value of the S channel by 20% as a whole, and convert the adjusted image back to the RGB space.

[0062] Analyze the image noise characteristics to determine whether it is Gaussian noise, salt-and-pepper noise, or other types. For Gaussian noise, use Gaussian filtering or bilateral filtering; for salt-and-pepper noise, use median filtering; for complex noise, use advanced noise reduction algorithms such as non-local means (NLM) or BM3D. Adjust the parameters of the noise reduction algorithm according to the image characteristics and noise level to achieve a balance between retaining details and removing noise.

[0063] Analyze the image brightness histogram to evaluate the degree of uneven illumination. Use histogram equalization or adaptive histogram equalization (CLAHE) for overall brightness adjustment. Apply local contrast enhancement algorithms such as the Retinex algorithm or multi-scale adaptive enhancement to identify the shadow and highlight areas in the image, perform brightness restoration and detail enhancement respectively, and analyze and correct the brightness difference in different facial areas caused by the light source position.

[0064] In this step, it also includes steps S13 - S15.

[0065] Step S13: Perform segmentation processing on the facial transition image after color correction, noise removal, and illumination equalization to obtain a facial transition segmentation map composed of multiple segmentation blocks.

[0066] In this step, use the 68-point facial feature detection model of the dlib library to locate 68 key feature points on the facial image, including eyes, eyebrows, nose, mouth, and facial contour. The face is divided into the following main regions according to the feature points: the area from the upper edge of the eyebrows to the hairline is divided into the forehead region, the left and right eyebrows are divided into the eyebrow region, the areas around the left and right eyes are divided into the eye region, the area from the center of the eyebrows to the tip of the nose is divided into the nose region, the left and right cheeks are divided into the cheek regions, the upper and lower lips are divided into the mouth region, and the area from the lower lip to the bottom of the chin is divided into the chin region.

[0067] Furthermore, the forehead region can be divided into 3 blocks horizontally and 2 blocks vertically, for a total of 6 segmentation blocks; the eyebrow region can divide each side of the eyebrows into 3 blocks, for a total of 6 segmentation blocks; the eye region can divide the area around each eye into 4 blocks, for a total of 8 segmentation blocks; the nose region can be divided into 3 blocks: upper, middle, and lower; the cheek region can divide each side of the cheeks into 4 blocks, for a total of 8 segmentation blocks; the mouth region can be divided into 2 blocks for the upper lip and 2 blocks for the lower lip, for a total of 4 segmentation blocks; the chin region can be divided into 3 blocks. Use the feature point coordinates to calculate the boundaries of each region. For example, the boundaries of the eye region can be determined by the feature points at the corners of the eyes and the lower edge of the eyebrows.

[0068] Create a blank mask image with the same size as the original image. Use the cv2.fillPoly( ) function to fill the corresponding polygon regions for each defined region and assign a unique identification number to each region. For the facial contour, use the feature points to fit a curve to ensure smooth segmentation edges. At key parts such as eyes and mouth, use finer segmentation to retain details.

[0069] Step S14: Obtain the single-channel grayscale image of each segmentation block in the facial transition segmentation map. According to the grayscale level standard, obtain the grayscale level of each pixel point in each segmentation block, and regard the pixel points with grayscale levels lower than the preset level threshold as under-enhanced pixel points.

[0070] Step S15: Adjust the pixel values of the pixel points corresponding to the under-enhanced pixel points in the facial transition image to the corresponding expected pixel values to obtain a facial standard image, where the expected pixel value is obtained based on the grayscale level of the corresponding pixel point, the total number of pixel values in the segmentation block where the corresponding pixel point is located, and the total number of grayscale levels.

[0071] In one example, assume that the right cheek area of a facial image is divided into a segmentation block of 50x50 pixels. The size of the segmentation block is 50x50 pixels, the total number of pixels (TP) is 2,500, the total number of gray levels (TGL) is 10 levels, the preset level threshold is 4 levels (pixels below level 4 are considered under-enhanced), and the range of each gray level is 255 / 10 = 25.5.

[0072] In this segmentation block, a pixel point with RGB values of (70, 65, 60) is determined. Calculate the gray level through the formula Gray = 0.299R + 0.587G + 0.114B, Gray = 0.299 ×70 + 0.587 ×65 + 0.114 ×60 ≈ 66. Determine that the current gray level is 66 / 25.5 ≈ 2.59, rounded down to level 2. Since 2 < 4 (preset threshold), this pixel is identified as an under-enhanced pixel. Determine the target level, target level = current level + 2 = 2 + 2 = 4 (not exceeding the highest level 10). The minimum value of level 4 is (4 - 1) ×25.5 = 76.5, and the maximum value of level 4 is 4 ×25.5 = 102. The relative position of the current value in level 2 is (66 - 25.5) / (51 - 25.5) ≈ 0.59. The expected value can be represented by E, E = 76.5 + 0.59 ×(102 - 76.5) ≈ 91.5. Adjust the original gray value 66 to the new expected value 92 (rounded), keeping the color ratio unchanged. The new RGB values are approximately R: 70 ×(92 / 66) ≈ 97, G: 65 ×(92 / 66) ≈ 90, B: 60 ×(92 / 66) ≈ 83. In the facial transition image, update the value of this pixel point from (70, 65, 60) to (97, 90, 83), and continue to repeat this process for all pixels identified as under-enhanced in this segmentation block.

[0073] Step S2: Based on the image feature information in the skin treatment segmentation dataset, construct a skin area division model. Input the facial standard image into the skin area division model for target recognition to obtain multiple images of areas to be treated in the facial standard image.

[0074] In this step, steps S21 to S24 are also included.

[0075] Step S21: Obtain the unimodal image data and multimodal image data in the skin treatment segmentation dataset, divide the unimodal image data and multimodal image data into first image data and second image data according to a preset ratio, and extract the data features in the first image data and second image data based on the feature extraction function.

[0076] In this step, the skin treatment segmentation dataset can be a large-scale skin lesion image dataset provided by the International Skin Imaging Collaboration (ISIC). Use Python's image processing libraries (such as PIL, OpenCV) to read the image files in the skin treatment segmentation dataset, traverse the dataset directory, and load the unimodal and multimodal images respectively.

[0077] Step S22: Based on the data features in the first image data, construct an initial multimodal attention fusion network, and use multi-scale feature fusion to increase the receptive field of the initial multimodal attention fusion network.

[0078] In this step, by constructing a two-branch network structure, process the RGB image and the depth image respectively. Each branch contains a feature extraction, an attention mechanism, and a multi-scale feature extraction module. Finally, the features of the two branches are combined through a fusion module. For the RGB branch, use the pre-trained ResNet50 as the backbone network. For the depth map branch, use the modified ResNet50 (the number of input channels is changed to 1). Remove the last fully connected layer of the two ResNet50s and retain the feature map output. Add a spatial attention module after the output of ResNet50 in each branch. Use a 1x1 convolutional layer to compress the feature map to a single channel, apply the Sigmoid activation function to obtain the attention weight, and multiply the attention weight with the original feature map. Design a multi-scale feature extraction module in each branch. Use 1x1, 3x3, and 5x5 convolutional layers to process the feature map in parallel. Use dilated convolution (atrous convolution) to increase the receptive field, with dilation rates of 1, 2, and 4 respectively, and splice the feature maps of different scales. Splice the multi-scale features of the RGB branch and the depth branch, use a 1x1 convolutional layer to reduce the number of feature channels, and apply global average pooling to obtain a fixed-length feature vector. Use the cross-entropy loss function. Adopt the Adam optimizer, set the initial learning rate to 0.001, and use a learning rate scheduler to train for 200 epochs, reducing the learning rate every 50 epochs.

[0079] Step S23: Extract the image depth features of the initial multimodal attention fusion network, and use the attention mechanism to calibrate the image depth feature channels in the initial multimodal attention fusion network to obtain the calibrated image depth features.

[0080] In this step, a channel attention module is added to the feature map after ResNet50 feature extraction. The feature map is processed using global average pooling and global max pooling respectively to obtain two channel descriptors. The two descriptors are processed through a shared multi-layer perceptron (MLP), which contains two fully-connected layers with a ReLU activation function in the middle. The outputs of the MLP are added together and passed through a Sigmoid function to obtain the channel attention weights. The obtained channel attention weights are multiplied by the original feature map to achieve importance weighting for different channels. After the channel attention, a spatial attention module is added. Max pooling and average pooling are performed separately along the channel dimension to obtain two 2D feature maps. The two 2D feature maps are concatenated and processed through a 7x7 convolutional layer, and a Sigmoid function is used to obtain the spatial attention weight map. The spatial attention weight map is multiplied by the feature map processed by the channel attention. The processed RGB features and depth features are concatenated, and a 1x1 convolutional layer is used to reduce the number of feature channels. A multi-scale feature extraction module is applied to the fused features, including convolutions with different kernel sizes and dilated convolutions with different dilation rates. The multi-scale features are concatenated, and a 1x1 convolution is used again to reduce the number of channels. Global average pooling is applied to obtain a fixed-length feature vector.

[0081] Step S24: Fuse the calibrated image depth features with the original facial image to construct a multi-modal attention fusion network, and train the multi-modal attention fusion network according to the data features and loss function in the second image data to obtain a skin area segmentation model.

[0082] In this embodiment, the expressions of the feature extraction function and the loss function are respectively:

[0083] ;

[0084] ;

[0085] Among them, M c ( F ) represents the feature extraction function, F represents the feature map input to the attention mechanism module, σ ( ) represents the activation function, MLP ( ) represents the optimization function built into the three-layer perceptron, AvgPool ( ) represents the average pooling function, W 0 represents the first weight matrix, W 1 represents the second weight matrix, represents the feature after average pooling, L total represents the loss function, α represents the weight loss parameter, qRepresents the true label value of the model output, p Represents the predicted value of the model output, δ Represents the edge adjustment parameter, N Represents the total number of samples input to the model, n Represents the n th sample, ω Represents the hyperparameter, q n Represents the n true label value of the p n Represents the n predicted value of the

[0086] Step S3: Perform independent component analysis on the image of the area to be treated to obtain the melanin channel and hemoglobin channel in the image of the area to be treated.

[0087] In this step, steps S31 to S32 are also included.

[0088] Step S31: Based on the skin color density function, construct a skin optical model, and introduce the skin optical model and the predicted pure color density vector into the skin optical model to obtain the synthetic skin color density function.

[0089] In this embodiment, the expression of the synthetic skin color density function is:

[0090] ;

[0091] ;

[0092] Wherein, I x,y represents the skin color density function, x represents the abscissa of the pixel point, y represents the abscissa of the pixel point, c m represents the pure color feature of melanin per unit density, c h represents the pure color feature of hemoglobin per unit density, represents the relative amount of melanin in each pixel point, represents the relative amount of hemoglobin in each pixel point, Δ h represents the stationary column vector generated by other skin pigments and structures, S x,y represents the synthetic skin color density function, represents the predicted pure color density vector, K represents the first density control parameter, j represents the second density control parameter.

[0093] Step S32: Based on the synthetic skin color density function and the least squares method, solve the melanin channel and the hemoglobin channel in the image of the area to be treated respectively.

[0094] In one example, assume that the skin includes three skin clusters: normal skin, pigmented skin, and erythema. A reference value vector is learned for each skin cluster by selecting a set of pixels from the image. To learn the reference values of the three skin clusters, it is required to manually outline the areas of normal skin, pigmented skin, and the vascular system in the selected training pixels. Perform independent component analysis on the selected image to extract the hemoglobin channel. Subsequently, calculate the mean and standard deviation of the R, G, and B channels of the hemoglobin components of normal skin, pigmented skin, and erythema. Store the obtained reference values as prior information, design a threshold framework based on the Mahalanobis distance on the skin hemoglobin components, and classify the red regions. For each pixel of the image, calculate the Mahalanobis distance of the three skin clusters (normal skin, pigmented skin, and blood vessels) in the hemoglobin component, and classify the pixel into the group with the closest distance.

[0095]

[0096] Among them, , , represent the Mahalanobis distances of the pixel (x, y) from normal skin, pigmented skin, and blood vessels respectively, and the subscripts ri , gi , bi are the RGB values of the reference clusters. If , then the pixel belongs to the erythema cluster. By performing the same analysis on each pixel of each image, each pixel is classified into one of the three mentioned skin clusters to achieve the segmentation of the skin area.

[0097] Furthermore, shape information can be considered while taking color into account, mainly considering two shape categories, namely tubular and circular skin blood vessels. To measure the tubularity of a pixel X = ([[]] x , y ) at different scales s = { s 1, s 2,..., s k} can be expressed using the following formula:

[0098]

[0099]

[0100]

[0101] Among them, R and S are measures of spot degree and second-order structure degree, λ i ( X , s ), i = 1, 2 (| λ 1| ≤ | λ 2|) is i the eigenvalue of the Hessian matrix of the image at scale s , β and c are control parameters that control the sensitivity of the filter to R and S . R reaches its maximum value at the clump structure, while S is very low in the background where there is no structure and the eigenvalues are small. The presence of tubular structures increases the contrast, and this function represents the class probability estimate of mapping the image to a tube.

[0102] According to the direction of subcutaneous blood vessels, some skin blood vessels may have a circular structure (dots). For a pixel to belong to a circular structure, the eigenvalues of the two Hessian matrices should be of the same order of magnitude. Therefore, the class probability estimate of mapping the image to a circle can be expressed as:

[0103]

[0104] Using the melanin channel and hemoglobin channel generated by the segmentation step, 12 vascular features are defined and extracted for each pathology in the historical skin treatment segmentation dataset. These features include the maximum vascular length, average vascular length, standard deviation of length, maximum vascular area, average vascular area, standard deviation of vascular area, ratio of vascular area to lesion area, maximum vascular width, average vascular width, standard deviation of vascular width, number of vascular branches, and ratio of branches to lesion area.

[0105] Step S4, construct a multi-modal attention network model, and input the melanin channel and hemoglobin channel into the multi-modal attention network model for feature extraction and segmentation, respectively obtaining melanin segmentation information and hemoglobin segmentation information.

[0106] In this embodiment, independent input branches are designed for the melanin channel and the hemoglobin channel respectively. This separate initial processing allows the model to capture the unique features of each channel respectively, laying a foundation for subsequent fusion and analysis. Each branch is equipped with a feature extraction network. For example, ResNet50 or DenseNet is used as the backbone network. These deep convolutional neural networks can extract rich multi-scale features from their respective channels, including low-level texture information and high-level semantic information. After feature extraction, multi-level attention mechanisms are introduced, including channel attention mechanism and spatial attention mechanism. The channel attention mechanism can adaptively adjust the importance of different feature maps, and the spatial attention mechanism is used to highlight the key regions in the image. These attention mechanisms are not only applied within a single channel, but also extended across channels to form a cross-modal attention layer, so as to establish a dynamic connection between melanin and hemoglobin information and capture the complex interactions between them.

[0107] Furthermore, the U-Net architecture is adopted as the decoder network. The U-Net architecture contains multiple upsampling blocks and fine-grained skip connections. These skip connections are not just simple feature transmissions, but are attention-weighted to ensure that only the most relevant low-level features are transmitted to the corresponding layers of the decoder, improving the model's ability to retain and utilize detailed information. The output layer of the multi-modal attention network model designs independent but interrelated output heads for melanin segmentation and hemoglobin segmentation respectively. Each output head not only contains a series of convolutional layers but also introduces a feedback mechanism, allowing information exchange between the two segmentation tasks.

[0108] In this embodiment, the melanin channel segmentation results include pigmented areas, which indicate areas with higher melanin content in the skin; hypopigmented areas, which identify areas with lower melanin content; pigment distribution patterns, which show the distribution of melanin in the skin, such as uniform distribution or irregular distribution; boundary information, which shows the boundary contours of pigmentary treatment; and structural features, such as specific melanin distribution patterns like reticular structure and globular structure. The hemoglobin channel segmentation results include vascular structures, which show the vascular network and distribution in the skin; erythematous areas, which identify areas with blood filling or inflammation in the skin; vascular morphology, such as different morphological vascular structures like linear, punctate, dendritic, etc.; vascular density, which shows the vascular density in different areas; and abnormal vascular structures, such as dilated blood vessels or abnormal vascular patterns. These segmentation results are usually presented in the form of binary images or multi-value label maps.

[0109] Step S5: Fuse the melanin segmentation information and hemoglobin segmentation information to obtain a skin treatment segmentation map, collect the repair features in the skin treatment segmentation map, classify the repair features according to a random forest classifier to obtain the skin treatment type corresponding to each image of the area to be treated, and give a treatment plan corresponding to the skin treatment type.

[0110] In this step, different weights are assigned to the melanin and hemoglobin segmentation results, and the weights can be determined based on clinical importance or specific treatment goals. For example, the final segmentation = α Melanin segmentation information + β Hemoglobin segmentation information, where α + β = 1. Create a multi-level classification system to reflect different treatment needs. Use color coding to visualize different treatment need areas. For example, green represents normal, yellow represents mild, orange represents moderate, and red represents severe. And use morphological operations (such as dilation and erosion) to enhance the segmentation boundaries, perform connected region analysis, identify and label independent treatment areas.

[0111] In one example, extract the melanin segmentation map and hemoglobin segmentation map from the melanin segmentation information and hemoglobin segmentation information, and the pixel value ranges of both are 0 - 255. Normalize the two segmentation maps to the range of 0 - 1, adjust the images to the same size to ensure that the two images have the same dimensions, assign weights to melanin and hemoglobin. For example, the melanin weight α = 0.6 and the hemoglobin weight β = 0.4, and the weight assignment can be adjusted according to the specific skin lesion type and clinical importance. Set thresholds to divide the fusion result into multiple levels. Among them, level 0 (normal) fusion value < 0.3; level 1 (mild) 0.3 ≤ fusion value < 0.5; level 2 (moderate) 0.5 ≤ fusion value < 0.7; level 3 (severe) fusion value ≥ 0.7. Represent different levels with different colors, level 0 is green (RGB: 0, 255, 0); level 1 is yellow (RGB: 255, 255, 0); level 2 is orange (RGB: 255, 165, 0); level 3 is red (RGB: 255, 0, 0). Use morphological operations (such as dilation) to enhance the boundaries of each area, perform connected region analysis on the fused image, assign a unique identifier to each independent area, calculate the total area and relative proportion of each level, and calculate the total area proportion that needs to be treated (levels 1 - 3).

[0112] In this embodiment, the random forest classifier includes a feature integration unit, a data set division unit, a decision tree construction unit, a multi - decision tree training unit, and a classification prediction unit, where,

[0113] The feature integration unit is used to integrate the repair features collected from the skin treatment segmentation map and form a repair feature dataset; the dataset division unit is used to divide the repair feature dataset into several different repair feature sub-datasets by random sampling; the decision tree construction unit is used to generate and recursively split the decision tree; the multi-decision tree training unit is used to build different decision trees on different repair feature sub-datasets respectively; the classification prediction unit is used to classify the samples and output the skin treatment type corresponding to each image of the area to be treated.

[0114] The skin treatment types include pigmentary diseases, vascular diseases, keratosis pilaris, acne and its sequelae, and skin texture problems.

[0115] By collecting the original facial image and combining it with the historical treatment dataset to construct a skin area division model, multiple areas to be treated on the face can be accurately identified and divided. At the same time, independent component analysis is performed on the areas to be treated to separate the melanin channel and the hemoglobin channel, which can analyze the pigmentation and vascular characteristics of the skin more deeply. The multi-modal attention network model is used to extract and segment features from the melanin and hemoglobin channels, which can consider both pigment and vascular features simultaneously. This multi-modal analysis can provide more comprehensive diagnostic information, thereby improving the accuracy of segmentation. And it can further optimize the features extracted by independent component analysis by taking advantage of the long-term dependence modeling of the skin area division model to complete the task of segmenting the images of the areas to be treated. Combining the advantages of the skin area division model and independent component analysis helps to improve the local detail sensitivity and global information processing efficiency of the skin area division model, so as to understand the image content more comprehensively.

[0116] Based on the above method, an intense pulsed light skin beauty local area recognition system is disclosed in an embodiment of the present application. Refer to Figure 2 , the local area recognition system 1 includes a data acquisition module 11, an information extraction module 12, and a treatment operation module 13. Among them,

[0117] The data acquisition module 11 is used to collect the original facial image of the patient and obtain the historical skin treatment segmentation dataset, and preprocess the original facial image to obtain the standard facial image;

[0118] The information extraction module 12 is used to construct a skin area division model based on the image feature information in the skin treatment segmentation dataset, input the standard facial image into the skin area division model for target recognition, obtain multiple images of the areas to be treated in the standard facial image, perform independent component analysis on the images of the areas to be treated, obtain the melanin channel and the hemoglobin channel in the images of the areas to be treated, construct a multi-modal attention network model, and input the melanin channel and the hemoglobin channel into the multi-modal attention network model for feature extraction and segmentation, and respectively obtain the melanin segmentation information and the hemoglobin segmentation information;

[0119] The treatment operation module 13 is used to fuse the melanin segmentation information and the hemoglobin segmentation information to obtain a skin treatment segmentation map, collect the repair features in the skin treatment segmentation map, classify the repair features according to a random forest classifier, obtain the skin treatment type corresponding to each image of the area to be treated, and give a treatment plan corresponding to the skin treatment type.

[0120] In one example, the data acquisition module 11 is used to convert the original facial image into the RBG format, adjust the facial area in the original facial image to a horizontal state to obtain a facial transition image; perform color correction, noise removal, illumination equalization, and detail enhancement processing on the facial transition image in sequence to obtain a standard facial image.

[0121] In one example, the data acquisition module 11 is used to perform segmentation processing on the facial transition image after color correction, noise removal, and illumination equalization processing to obtain a facial transition segmentation map composed of multiple segmentation blocks; obtain the single-channel grayscale image of each segmentation block in the facial transition segmentation map, obtain the gray level of each pixel point in each segmentation block according to the gray level standard, and use the pixel points with gray levels lower than the preset level threshold as under-enhanced pixel points; adjust the pixel values of the pixel points corresponding to the under-enhanced pixel points in the facial transition image to the corresponding expected pixel values to obtain a standard facial image, where the expected pixel value is obtained based on the gray level of the corresponding pixel point, the total number of pixel values of the segmentation block where the corresponding pixel point is located, and the total number of gray levels.

[0122] In one example, the information extraction module 12 is used to obtain the single-modal image data and multi-modal image data in the skin treatment segmentation dataset, divide the single-modal image data and multi-modal image data into first image data and second image data at a preset ratio, and extract the data features in the first image data and second image data based on a feature extraction function; based on the data features in the first image data, construct an initial multi-modal attention fusion network, and use multi-scale feature fusion to increase the receptive field of the initial multi-modal attention fusion network; perform image depth feature extraction on the initial multi-modal attention fusion network, and use an attention mechanism to calibrate the image depth feature channels in the initial multi-modal attention fusion network to obtain calibrated image depth features; fuse the calibrated image depth features with the original facial image to construct a multi-modal attention fusion network, and train the multi-modal attention fusion network according to the data features in the second image data and a loss function to obtain a skin area division model.

[0123] In one example, the expressions of the feature extraction function and the loss function are respectively:

[0124] ;

[0125] ;

[0126] Among them, M c ( F ) represents the feature extraction function, F represents the feature map input to the input attention mechanism module, σ ( ) represents the activation function, MLP ( ) represents the optimization function built into the three-layer perceptron, AvgPool ( ) represents the average pooling function, W 0 represents the first weight matrix, W 1 represents the second weight matrix, represents the feature after average pooling, L total represents the loss function, α represents the weight loss parameter, q represents the true label value output by the model, p represents the predicted value output by the model, δ represents the edge adjustment parameter, N represents the total number of samples input to the model, n represents the n th sample, ω represents the hyperparameter, q n represents the n th true label value of the sample, p n represents the n th predicted value of the sample.

[0127] In one example, the information extraction module 12 is used to construct a skin optical model based on the skin color density function, and introduce the skin optical model and the predicted pure color density vector into the skin optical model to obtain a synthetic skin color density function; based on the synthetic skin color density function and the least squares method, the melanin channel and the hemoglobin channel in the image of the area to be treated are solved respectively.

[0128] In one example, the expression of the synthetic skin color density function is:

[0129] ;

[0130] ;

[0131] Among them, I x,y represents the skin color density function, x represents the abscissa of the pixel point, y represents the abscissa of the pixel point, c mRepresents the pure color feature of melanin per unit density, c h Represents the pure color feature of hemoglobin per unit density, Represents the relative amount of melanin in each pixel, Represents the relative amount of hemoglobin in each pixel, Δ h Represents the stationary column vector generated by other skin pigments and structures, S x,y Represents the synthetic skin color density function, Represents the predicted pure color density vector, K Represents the first density control parameter, j Represents the second density control parameter.

[0132] In one example, the random forest classifier includes a feature integration unit, a data set division unit, a decision tree construction unit, a multi-decision tree training unit, and a classification prediction unit. Among them,

[0133] The feature integration unit is used to integrate the repair features collected from the skin treatment segmentation map and form a repair feature data set;

[0134] The data set division unit is used to divide the repair feature data set into several different repair feature sub-data sets by random sampling;

[0135] The decision tree construction unit is used to generate decision trees and perform recursive splitting;

[0136] The multi-decision tree training unit is used to establish different decision trees on different repair feature sub-data sets respectively;

[0137] The classification prediction unit is used to perform sample classification and output the skin treatment type corresponding to each image of the area to be treated.

[0138] In one example, the skin treatment types include pigmentary diseases, vascular diseases, keratosis pilaris, acne and its sequelae, and skin texture problems.

[0139] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for identifying a local area of intense pulsed light skin beauty, characterized in that, The method includes: Collect the original facial image of the patient and obtain the historical skin treatment segmentation dataset, preprocess the original facial image to obtain the standard facial image; Based on the image feature information in the skin treatment segmentation dataset, construct a skin area division model, input the standard facial image into the skin area division model for target recognition, and obtain multiple images of areas to be treated in the standard facial image; The constructing a skin area division model based on the image feature information in the skin treatment segmentation dataset specifically includes: Obtain the unimodal image data and multimodal image data in the skin treatment segmentation dataset, divide the unimodal image data and multimodal image data into first image data and second image data in a preset ratio, and extract the data features in the first image data and the second image data based on the feature extraction function; Based on the data features in the first image data, construct an initial multimodal attention fusion network, and use multi-scale feature fusion to increase the receptive field of the initial multimodal attention fusion network; Extract the image depth features of the initial multimodal attention fusion network, and use the attention mechanism to calibrate the image depth feature channels in the initial multimodal attention fusion network to obtain the calibrated image depth features; Fuse the calibrated image depth features with the original facial image to construct a multimodal attention fusion network, and train the multimodal attention fusion network according to the data features and loss function in the second image data to obtain the skin area division model; The expressions of the feature extraction function and the loss function are respectively: ； ； Among them, M c ( F ) represents the feature extraction function, F represents the feature map input to the input attention mechanism module, σ ( ) represents the activation function, MLP ( ) represents the optimization function built into the three-layer perceptron, AvgPool ( ) represents the average pooling function, W 0 represents the first weight matrix, W 1 represents the second weight matrix, represents the feature after average pooling, L total represents the loss function, α represents the weight loss parameter, q represents the true label value output by the model, p represents the predicted value output by the model, δ represents the edge adjustment parameter, N represents the total number of samples input to the model, n represents the n th sample, ω represents the hyperparameter, q n represents the n th true label value of the sample, p n represents the n th predicted value of the sample; Perform independent component analysis on the images of areas to be treated to obtain the melanin channel and hemoglobin channel in the images of areas to be treated; The performing independent component analysis on the images of areas to be treated to obtain the melanin channel and hemoglobin channel in the images of areas to be treated specifically includes: Based on the skin color density function, construct a skin optical model, and introduce the skin optical model and the predicted pure color density vector into the skin optical model to obtain the synthetic skin color density function; Based on the synthetic skin color density function and the least squares method, solve the melanin channel and hemoglobin channel in the images of areas to be treated respectively; The expression of the synthetic skin color density function is: ； ； Among them, I x,y represents the skin color density function, x represents the abscissa of the pixel point, y represents the abscissa of the pixel point, c m represents the pure color feature of melanin per unit density, c h represents the pure color feature of hemoglobin per unit density, represents the relative amount of melanin in each pixel point, represents the relative amount of hemoglobin in each pixel point, Δ h represents the stationary column vector generated by other skin pigments and structures, S x,y represents the synthetic skin color density function, represents the predicted pure color density vector, K represents the first density control parameter, j represents the second density control parameter; Construct a multimodal attention network model, input the melanin channel and the hemoglobin channel into the multimodal attention network model for feature extraction and segmentation, and obtain the melanin segmentation information and hemoglobin segmentation information respectively; Fuse the melanin segmentation information and the hemoglobin segmentation information to obtain a skin treatment segmentation map, collect the repair features in the skin treatment segmentation map, classify the repair features according to the random forest classifier, obtain the skin treatment type corresponding to each image of the area to be treated, and give a treatment plan corresponding to the skin treatment type.

2. The method according to claim 1, characterized in that, The preprocessing the original facial image to obtain the standard facial image specifically includes: Convert the original facial image to the RBG format and adjust the facial area in the original facial image to a horizontal state to obtain a facial transition image; Perform color correction, noise removal, illumination equalization, and detail enhancement processing on the facial transition image in sequence to obtain a facial standard image.

3. The method according to claim 2, characterized in that, The detail enhancement specifically includes: Perform segmentation processing on the facial transition image after color correction, noise removal, and illumination equalization processing to obtain a facial transition segmentation map composed of multiple segmentation blocks; Obtain the single-channel grayscale image of each segmentation block in the facial transition segmentation map, obtain the gray level of each pixel point in each segmentation block according to the gray level standard, and use the pixel points with gray levels lower than the preset level threshold as under-enhanced pixel points; Adjust the pixel value of the pixel point corresponding to the under-enhanced pixel point in the facial transition image to the corresponding expected pixel value to obtain the facial standard image, where the expected pixel value is obtained based on the gray level of the corresponding pixel point, the total number of pixel values of the segmentation block where the corresponding pixel point is located, and the total number of gray levels.

4. The method according to claim 1, wherein The random forest classifier includes a feature integration unit, a data set division unit, a decision tree construction unit, a multi-decision tree training unit, and a classification prediction unit, where The feature integration unit is used to integrate the repair features collected in the skin treatment segmentation map and form a repair feature data set; The data set division unit is used to divide the repair feature data set into several different repair feature sub-data sets by random sampling; The decision tree construction unit is used to generate a decision tree and perform recursive splitting; The multi-decision tree training unit is used to establish different decision trees on different repair feature sub-data sets respectively; The classification prediction unit is used to perform sample classification and output the skin treatment type corresponding to each image of the area to be treated.

5. The method according to claim 1, characterized in that, The skin treatment types include pigmentary diseases, vascular diseases, keratosis pilaris, acne and its sequelae, and skin texture problems.

6. A local area recognition system for intense pulsed light skin beauty, characterized in that, The local area recognition system (1) includes a data acquisition module (11), an information extraction module (12), and a treatment operation module (13), where The data acquisition module (11) is used to acquire the original facial image of the patient and obtain the historical skin treatment segmentation data set, preprocess the original facial image, and obtain a facial standard image; The information extraction module (12) is used to construct a skin area division model based on the image feature information in the skin treatment segmentation data set, input the facial standard image into the skin area division model for target recognition, obtain multiple images of the areas to be treated in the facial standard image, perform independent component analysis on the images of the areas to be treated, obtain the melanin channel and hemoglobin channel in the images of the areas to be treated, construct a multi-modal attention network model, and input the melanin channel and the hemoglobin channel into the multi-modal attention network model for feature extraction and segmentation to obtain melanin segmentation information and hemoglobin segmentation information respectively; Constructing a skin area division model based on the image feature information in the skin treatment segmentation data set specifically includes: Obtain unimodal image data and multimodal image data in the skin treatment segmentation dataset, divide the unimodal image data and multimodal image data into first image data and second image data at a preset ratio, and extract data features in the first image data and the second image data based on a feature extraction function; Based on the data features in the first image data, construct an initial multimodal attention fusion network, and use multi-scale feature fusion to increase the receptive field of the initial multimodal attention fusion network; Perform image depth feature extraction on the initial multimodal attention fusion network, and use an attention mechanism to calibrate the image depth feature channels in the initial multimodal attention fusion network to obtain calibrated image depth features; Fuse the calibrated image depth features with the original facial image to construct a multimodal attention fusion network, and train the multimodal attention fusion network according to the data features and loss function in the second image data to obtain the skin area segmentation model; The expressions of the feature extraction function and the loss function are respectively: The expressions of the feature extraction function and the loss function are respectively: ； ； Among them, M c ( F ) represents the feature extraction function, F represents the feature map input to the input attention mechanism module, σ ( ) represents the activation function, MLP ( ) represents the optimization function built into the three-layer perceptron, AvgPool ( ) represents the average pooling function, W 0 represents the first weight matrix, W 1 represents the second weight matrix, represents the feature after average pooling, L total represents the loss function, α represents the weight loss parameter, q represents the true label value output by the model, p represents the predicted value output by the model, δ represents the edge adjustment parameter, N represents the total number of samples input to the model, n represents the n th sample, ω represents the hyperparameter, q n represents the n th true label value of the sample, p n represents the n th predicted value of the sample; Perform independent component analysis on the image of the area to be treated to obtain the melanin channel and hemoglobin channel in the image of the area to be treated, specifically including: Based on the skin color density function, construct a skin optical model, and introduce the skin optical model and the predicted pure color density vector into the skin optical model to obtain a synthetic skin color density function; Based on the synthetic skin color density function and the least squares method, solve the melanin channel and hemoglobin channel in the image of the area to be treated respectively; The expression of the synthetic skin color density function is: ； ； Among them, I x,y represents the skin color density function, x represents the abscissa of the pixel point, y represents the abscissa of the pixel point, c m represents the pure color feature of melanin per unit density, c h represents the pure color feature of hemoglobin per unit density, represents the relative amount of melanin in each pixel point, represents the relative amount of hemoglobin in each pixel point, Δ h represents the stationary column vector generated by other skin pigments and structures, S x,y represents the synthetic skin color density function, represents the predicted pure color density vector, K represents the first density control parameter, j represents the second density control parameter; The treatment operation module (13) is used to fuse the melanin segmentation information and the hemoglobin segmentation information to obtain a skin treatment segmentation map, collect the repair features in the skin treatment segmentation map, classify the repair features according to a random forest classifier, obtain the skin treatment type corresponding to each image of the area to be treated, and give a treatment plan corresponding to the skin treatment type.

Images