Facial blepharoma positioning and measuring method and device based on deep learning
Through the deep learning-based measurement method of facial palpitations, the deep learning image segmentation model and contour extraction analysis technology are used to solve the problem of low recognition accuracy in traditional methods under complex conditions, and more accurate and efficient palpitations recognition and area calculation are achieved.
Patent Information
- Application Number
- CN202411774182.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-05
- Publication Date
- 2025-05-06
AI Technical Summary
Traditional facial palpebral xanthoma recognition technology has low recognition accuracy under complex backgrounds and different skin types, and is prone to missed detection.
Using deep learning-based facial pallicular xanthoma positioning measurement method, facial images are collected through visual sensors, and after preprocessing, pixel-level segmentation is performed using a deep learning image segmentation model (such as CNN or U-Net), and a mask map marked with pallicular xanthoma area is output, and the actual area of pallicular xanthoma is calculated based on contour extraction and connectivity area analysis.
Under complex backgrounds, different skin tones and different lighting conditions, it is possible to identify and locate palpitations more accurately and efficiently, reducing errors, and improving adaptability and robustness.
Smart Images
Figure CN119941843A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical auxiliary technology for facial xanthelasma, and in particular to a method and device for positioning and measuring facial xanthelasma based on deep learning. Background Art
[0002] Xanthelasma palpebrarum is a common benign skin lesion that usually appears in the eyelid area. It manifests as yellow, plaque-like lipid deposits that affect appearance. Traditional methods for identifying xanthelasma palpebrarum rely on the doctor's visual judgment, which has the problems of subjectivity and limited recognition accuracy. With the development of medical imaging technology, the combination of computer vision and machine learning methods to locate and measure the area of facial xanthelasma palpebrarum provides new possibilities for assisting doctors in identifying xanthelasma palpebrarum, which can effectively help doctors improve the accuracy and efficiency of recognition. However, the existing xanthelasma palpebrarum recognition technology is mainly based on traditional image processing methods, which has certain limitations, especially in complex backgrounds and different skin types, the recognition accuracy is reduced. Since the appearance of xanthelasma palpebrarum varies greatly among different individuals and in different shooting environments, traditional image processing methods are prone to false detection or missed detection.
[0003] Based on this, the present invention is proposed. Summary of the invention
[0004] The purpose of the present invention is to provide a facial xanthelasma positioning and measurement method and device based on deep learning, which can more accurately and efficiently identify xanthelasma under complex backgrounds, different skin colors, different lighting conditions, etc.
[0005] To achieve the above object, the present invention provides the following technical solutions:
[0006] In a first aspect, the present invention provides a facial xanthoma positioning and measurement method based on deep learning, including: S100 image acquisition and preprocessing, acquiring facial images through a visual sensor, and preprocessing the acquired facial images; S200 xanthoma positioning: establishing and training a deep learning image segmentation model, inputting the preprocessed facial image into the trained deep learning image segmentation model for pixel-level segmentation, and outputting a mask image marking the xanthoma area; S300 xanthoma area calculation: calculating the actual area of the xanthoma area, specifically including the following steps: S310 contour extraction, extracting the boundary contour of the xanthoma area; S320 connected region analysis, for irregular shapes or overlapping areas, using a connected region analysis algorithm to determine the connectivity of the area; S330 area calculation, calculating the area of the xanthoma area based on the contour boundary and connected region analysis results.
[0007] To facilitate subsequent segmentation and positioning, the present invention provides a preferred solution in the first aspect. In step S100, image acquisition and preprocessing, the preprocessing includes S110 denoising, which performs denoising on the acquired facial image to remove random noise in the image.
[0008] In order to further improve adaptability and robustness and enhance accuracy, the present invention provides a preferred solution in the first aspect. In step S200, in xanthelasma locating, the deep learning image segmentation model adopts a CNN convolutional neural network model or a U-Net convolutional neural network model.
[0009] In order to further assist decision making, the present invention provides a preferred solution in the first aspect, wherein the positioning measurement method further includes: S400 result output and report generation, providing the positioning information and area calculation results of the xanthelasma area, and generating a report.
[0010] In a second aspect, the present invention provides a facial xanthelasma positioning and measurement device based on deep learning, which adopts the above-mentioned positioning and measurement method, including: an image acquisition and preprocessing module, which is used to acquire facial images through a visual sensor and preprocess the acquired facial images; a xanthelasma positioning module, which is used to establish and train a deep learning image segmentation model, input the preprocessed facial image into the trained deep learning image segmentation model for pixel-level segmentation, and output a mask image marking the xanthelasma area; a xanthelasma area calculation module, which is used to calculate the actual area of the xanthelasma area, and specifically includes the following submodules: a contour extraction submodule, which is used to extract the boundary contour of the xanthelasma area; a connected region analysis submodule, which is used to determine the connectivity of irregular shapes or overlapping areas using a connected region analysis algorithm; and an area calculation submodule, which is used to calculate the area of the xanthelasma area based on the contour boundary and connected region analysis results.
[0011] Compared with the prior art, the above technical solution has the following beneficial technical effects:
[0012] The facial xanthelasma positioning and measurement method and device based on deep learning of the present invention can not only identify xanthelasma more accurately and efficiently under complex backgrounds, different skin colors, different lighting conditions, etc., with good adaptability and robustness, but also can accurately locate xanthelasma at the pixel level, avoiding the positioning error caused by blurred edges or complex backgrounds in traditional methods. After obtaining the accurate boundary of the lesion area through the deep learning model, combined with the contour boundary and the connected area, the actual area of the xanthelasma can be accurately calculated, reducing the area error caused by inaccurate edge detection in traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0014] Figure 1 A flowchart of a facial xanthelasma positioning and measurement method based on deep learning provided in Example 1 of the present invention;
[0015] Figure 2 A specific flow chart of xanthoma area calculation in the facial xanthoma positioning and measurement method based on deep learning provided in Example 1 of the present invention;
[0016] Figure 3 A schematic diagram of a module of a facial xanthelasma positioning and measuring device based on deep learning provided in Example 1 of the present invention;
[0017] Figure 4 A flowchart of a facial xanthelasma positioning and measurement method based on deep learning provided in Example 2 of the present invention;
[0018] Figure 5 A specific flow chart of denoising processing in the facial xanthelasma positioning and measurement method based on deep learning provided in Example 2 of the present invention;
[0019] Figure 6 A schematic diagram of a module of a facial xanthelasma positioning and measuring device based on deep learning provided in Example 2 of the present invention;
[0020] Figure 7 Flow chart of the facial xanthelasma positioning and measurement method based on deep learning provided in Example 3 of the present invention:
[0021] Figure 8 This is a module schematic diagram of the facial xanthelasma positioning and measurement device based on deep learning provided in Example 3 of the present invention.
[0022] Figure numerals: image acquisition and preprocessing module 100, image acquisition submodule 110, denoising processing submodule 120, Gaussian filtering submodule 121, median filtering submodule 122, contrast enhancement submodule 123, color space conversion submodule 124, xanthelasma positioning module 200, post-processing module 210, xanthelasma area calculation module 300, contour extraction submodule 310, connected region analysis submodule 320, area calculation submodule 330, result output and report generation module 400. DETAILED DESCRIPTION
[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0024] Example 1
[0025] Please refer to Figure 1 This embodiment provides a facial xanthelasma positioning and measurement method based on deep learning, which is mainly implemented through the following steps:
[0026] S100 image acquisition and preprocessing, collects facial images through visual sensors and preprocesses the collected facial images. This step requires obtaining clear facial images and performing preliminary image processing to provide good input data for subsequent lesion area detection and area calculation.
[0027] S200 xanthelasma localization, establish and train a deep learning image segmentation model, input the preprocessed facial image into the trained deep learning image segmentation model for pixel-level segmentation, and output a mask image marking the xanthelasma area. This step accurately locates the xanthelasma area through image segmentation technology, laying the foundation for subsequent area calculation.
[0028] S300 xanthelasma area calculation, calculate the actual area of the xanthelasma area.
[0029] Please refer to Figure 2 , wherein step S300 specifically includes the following sub-steps:
[0030] S310 contour extraction, extracting the boundary contour of the xanthelasma area;
[0031] S320 Connected Region Analysis: For irregular shapes or overlapping regions, the connected region analysis algorithm is used to determine the connectivity of the region;
[0032] S330 area calculation, based on the contour boundary and connected area analysis results, calculates the area of the xanthelasma region, providing quantitative data support for auxiliary diagnosis and clinical applications.
[0033] Please refer to Figure 3The present embodiment provides a facial xanthelasma positioning and measuring device based on deep learning, which adopts the above positioning and measuring method and is mainly composed of the following modules: an image acquisition and preprocessing module 100, which is used to acquire facial images through a visual sensor and preprocess the acquired facial images; a xanthelasma positioning module 200, which is used to establish and train a deep learning image segmentation model, input the preprocessed facial image into the trained deep learning image segmentation model for pixel-level segmentation, and output a mask image marking the xanthelasma area; a xanthelasma area calculation module 300, which is used to calculate the actual area of the xanthelasma area, and the xanthelasma area calculation module 300 specifically includes the following submodules: a contour extraction submodule 310, which is used to extract the boundary contour of the xanthelasma area; a connected region analysis submodule 320, which is used to determine the connectivity of irregular shapes or overlapping areas by using a connected region analysis algorithm; an area calculation submodule 330, which is used to calculate the area of the xanthelasma area based on the contour boundary and the connected region analysis results.
[0034] The facial xanthelasma positioning and measurement method and device based on deep learning in this embodiment, by introducing a deep learning image segmentation model, can not only more accurately and efficiently identify xanthelasma under complex backgrounds, different skin colors, different lighting conditions, etc., with good adaptability and robustness, but also can accurately locate xanthelasma at the pixel level, avoiding the positioning error caused by blurred edges or complex backgrounds in traditional methods. After obtaining the accurate boundary of the lesion area through the deep learning model, combined with the contour boundary and the connected area, the actual area of the xanthelasma can be accurately calculated, reducing the area error caused by inaccurate edge detection in traditional methods.
[0035] Example 2
[0036] On the basis of Example 1, in the facial xanthelasma positioning and measurement method based on deep learning in this embodiment, a more detailed and preferred solution is given for the image acquisition and preprocessing in step S100, and a specific deep learning image segmentation model is given:
[0037] First, please refer to Figure 4 In the embodiment step S100 image acquisition and preprocessing, it includes S110 image acquisition, which acquires facial images through a visual sensor. The visual sensor uses a high-resolution camera (such as a digital camera or a front camera of a smart phone) to acquire facial images, ensuring that the acquired images are clear and have no obvious blur. To ensure the best effect, the shooting can be carried out under stable lighting conditions to avoid strong shadows or overexposure. S120 denoising processing, denoising the acquired facial images to remove random noise in the image.
[0038] Please refer to Figure 5 The denoising process in step S120 is performed in the following steps:
[0039] S111 Gaussian filter, smoothes facial images, removes Gaussian noise, and reduces the interference of straw ropes on subsequent processing.
[0040] S112 median filtering effectively removes salt and pepper noise and protects edge information.
[0041] S113 contrast enhancement uses histogram equalization or adaptive contrast enhancement methods to enhance image details and ensure that the contrast between the xanthelasma area and the background is high enough to facilitate subsequent segmentation and positioning.
[0042] S114 color space conversion, converting the image from RGB space to HSV space or Lab space, helps to improve the stability of the image under different lighting conditions, especially under complex skin colors and backgrounds. Of course, in some embodiments, the denoising process can also adopt any one or a combination of two or more of the above steps according to the actual recognition accuracy requirements.
[0043] Secondly, in step S200 of xanthelasma positioning, the deep learning image segmentation model adopts a CNN convolutional neural network model or a U-Net convolutional neural network model. First, a dataset of annotated xanthelasma images is collected, and the model is trained by annotated data so that it can automatically identify and segment the lesion area (xanthelasma area). CNN and U-Net image segmentation networks are particularly suitable for medical image segmentation tasks, and can be segmented accurately to the pixel level, thereby achieving accurate positioning of xanthelasma. The encoder part of U-Net extracts image features through convolution operations, and the decoder part restores the image resolution through deconvolution, and finally generates a mask of the lesion area. The trained deep learning model performs pixel-level prediction on the input facial image and marks the area of xanthelasma. In this step, the model outputs a mask image of the same size as the input image, indicating whether each pixel belongs to the xanthelasma area.
[0044] Please refer to Figure 6 The present embodiment provides a facial xanthelasma positioning and measuring device based on deep learning, wherein the image acquisition and preprocessing module 100 includes an image acquisition submodule 110 and a denoising processing submodule 120, and the denoising processing submodule 120 is composed of a Gaussian filter submodule 121, a median filter submodule 122, a contrast enhancement submodule 123 and a color space conversion submodule 124 to execute the xanthelasma positioning and measuring method of the present embodiment.
[0045] This embodiment can further optimize the facial image and improve the accuracy of xanthelasma positioning measurement through multiple denoising processing steps.
[0046] Example 3
[0047] On the basis of Example 1 or 2, a post-processing step and a result output step are added, and a method for contour extraction and a specific area calculation method are further provided.
[0048] Step S310 contour extraction can identify the boundary of the xanthelasma region by using the Canny edge detection algorithm. This method has a strong ability to identify the edge of the image, and is particularly suitable for situations where the contours in the image are relatively clear. The boundary of the xanthelasma region can also be extracted by performing Hough transform on the xanthelasma region to detect straight lines or circles, and the contour extraction result can be further optimized. A region growing algorithm can also be used, that is, starting from a seed point in the xanthelasma region, the region growing algorithm is used to determine the entire xanthelasma region and obtain the boundary of the xanthelasma region, which is suitable for irregularly shaped regions.
[0049] Step S330 area calculation: based on the contour boundary and connected area analysis results, the area of the xanthelasma region is calculated. The area of the xanthelasma region can be obtained by counting the number of pixels in the xanthelasma region and converting it according to the resolution of the image (for example, the relationship between the pixel size and the actual area). The calculated area value can be used as a basis for diagnosis to help doctors assess the severity of the lesion.
[0050] Please refer to Figure 7 After the xanthelasma location is completed in step S200 and before the xanthelasma area is calculated in step S300, step S210 post-processing is performed, and the segmented xanthelasma region is further optimized by morphological operations. After the preliminary lesion region (xanthelasma region) is segmented, the segmentation result can be further optimized by morphological operations (such as corrosion and expansion) to remove small noise regions and fill small holes in the lesion region to ensure the consistency and integrity of the segmentation result.
[0051] After the xanthelasma area calculation in step S300, a step S400 is added for result output and report generation, which provides the location information and area calculation results of the xanthelasma area and generates a report. The location information and area calculation results of the lesion area (xanthelasma area) are provided to the doctor, and an automated auxiliary diagnosis report is generated to assist clinical decision-making. Further, this can be achieved through the following steps:
[0052] Image annotation: The segmented xanthelasma area is annotated on the original image, and the lesion area is highlighted with different colors or borders. In this way, the doctor can clearly see the location and range of the lesion.
[0053] Report generation: Based on the location, area and other information of the lesion area, an auxiliary diagnosis report is automatically generated. The report mainly includes the regional location and size of the xanthelasma, the calculated area value, and comparison with the normal skin area, preliminary assessment suggestions based on the area (such as whether further examination or treatment is needed), etc.
[0054] Result display: The annotated images and reports are presented to doctors in a visual form. The reports can be exported in PDF or image file format for easy archiving and sharing.
[0055] Please refer to Figure 8 The present embodiment provides a facial xanthelasma positioning and measuring device based on deep learning, which specifically includes an image acquisition and preprocessing module 100, a xanthelasma positioning module 200, a post-processing module 210, a xanthelasma area calculation module 300 and a result output and report generation module 400.
[0056] To further support the above positioning measurement methods and devices, system integration and user interaction can be set up to provide clinicians with an easy-to-use operating interface, simplify the auxiliary diagnosis process, and improve the efficiency of auxiliary diagnosis. The implementation schemes that can be adopted are as follows: User interface design: Design a simple and intuitive user interface. Doctors can diagnose by uploading facial images or taking images through real-time cameras. The interface should include functions such as image uploading, processing progress, and result display. Automation of the processing process: Once the image is uploaded, the system will automatically perform preprocessing, segmentation, area calculation, and report generation. The doctor only needs to make a final diagnosis confirmation. Feedback and adjustment: If the model fails to correctly identify the lesion area or calculate the area, the doctor can manually adjust the segmented area, and the system will update according to the adjustment and generate a new report.
[0057] Based on the above embodiments, the facial xanthelasma positioning and measurement method and device based on deep learning of the present invention can mainly achieve the following beneficial technical effects:
[0058] (1) Improved adaptability and robustness: By introducing deep learning technology, especially convolutional neural network (CNN) and other models, the present invention can accurately identify xanthelasma under complex backgrounds, different skin colors, different lighting conditions, etc., and has stronger universality and robustness. Compared with traditional methods, the present invention can accurately locate xanthelasma under various shooting conditions and calculate its area, reducing the possibility of false detection and missed detection.
[0059] (2) Accurately locate the lesion area: Advanced image segmentation technology (such as U-Net or other deep neural networks) can be used to accurately locate xanthelasma at the pixel level, avoiding the positioning errors caused by blurred edges or complex backgrounds in traditional methods.
[0060] (3) Accurate area calculation: After obtaining the accurate boundary of the lesion area through the deep learning model, combined with algorithms such as morphological analysis and connected region analysis, the actual area of the xanthelasma can be accurately calculated. This can cope with complex situations such as irregular morphology and partial occlusion of xanthelasma, thus ensuring the accuracy of area measurement.
[0061] (4) Strong anti-noise ability: The present invention can effectively remove noise in the image (such as lighting changes, skin texture, etc.) through the feature extraction capability of the deep learning model, improve the stability and accuracy of the processing, and obtain reliable results even when there is a certain amount of noise in the image.
[0062] (5) The entire process from image acquisition to result output can be automated, reducing the workload of doctors and improving diagnostic efficiency.
[0063] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.
[0064] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0065] The above-described embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.
Claims
1. A facial xanthelasma positioning and measurement method based on deep learning, characterized in that: include: S100 image acquisition and preprocessing, which uses visual sensors to acquire facial images and preprocess the acquired facial images; S200 xanthelasma localization, establish and train a deep learning image segmentation model, input the preprocessed facial image into the trained deep learning image segmentation model for pixel-level segmentation, and output a mask image marking the xanthelasma area; S300 Xanthelasma Area Calculation: Calculate the actual area of the xanthelasma area, including the following steps: S310 contour extraction, extracting the boundary contour of the xanthelasma area; S320 Connected Region Analysis: For irregular shapes or overlapping regions, the connected region analysis algorithm is used to determine the connectivity of the region; S330 area calculation, based on the contour boundary and connected area analysis results, calculates the area of the xanthelasma region.
2. The facial xanthelasma positioning and measurement method based on deep learning according to claim 1 is characterized in that: Step S100 of image acquisition and preprocessing includes step S110 of image acquisition and step S120 of denoising, in which the acquired facial image is denoised to remove random noise points in the image.
3. The facial xanthelasma positioning and measurement method based on deep learning according to claim 2 is characterized in that: The denoising process adopts any one or a combination of two or more of the following steps: S111 Gaussian filtering to remove Gaussian noise; S112 median filtering to remove salt and pepper noise; S113 contrast enhancement to enhance the details of the image using histogram equalization or adaptive contrast enhancement method; S114 color space conversion to convert the image from RGB space to HSV space or Lab space.
4. The facial xanthelasma positioning and measurement method based on deep learning according to claim 1 is characterized in that: In step S200 of xanthelasma locating, the deep learning image segmentation model adopts a CNN convolutional neural network model or a U-Net convolutional neural network model.
5. The facial xanthelasma positioning and measurement method based on deep learning according to claim 1 is characterized in that: After the xanthelasma positioning is completed in step S200 and before the xanthelasma area calculation in step S300, a post-processing step is performed: the segmented xanthelasma area is further optimized through morphological operations.
6. The facial xanthelasma positioning and measurement method based on deep learning according to claim 1, characterized in that: Step S310 is contour extraction, in which the boundary of the xanthelasma region is identified by using the Canny edge detection algorithm.
7. The facial xanthelasma positioning and measurement method based on deep learning according to claim 1 is characterized in that: Step S310 is contour extraction, which extracts the boundary of the xanthelasma region by performing Hough transform of straight line or circle detection on the xanthelasma region, or, starting from a seed point of the xanthelasma region, uses a region growing algorithm to determine the entire xanthelasma region to obtain the boundary of the xanthelasma region.
8. The facial xanthelasma positioning and measurement method based on deep learning according to claim 1, characterized in that: Step S330 area calculation: by counting the number of pixels in the xanthelasma region and converting it according to the resolution of the image, the area of the xanthelasma region is obtained.
9. The facial xanthelasma positioning and measurement method based on deep learning according to claim 1, characterized in that: Also includes: S400 result output and report generation, giving the location information of the xanthelasma area, area calculation results, and generating a report.
10. A facial xanthelasma positioning and measuring device based on deep learning, using the positioning and measuring method according to any one of claims 1 to 9, characterized in that: include: An image acquisition and preprocessing module is used to acquire facial images through a visual sensor and preprocess the acquired facial images; The xanthelasma localization module is used to establish and train a deep learning image segmentation model, input the preprocessed facial image into the trained deep learning image segmentation model for pixel-level segmentation, and output a mask image marking the xanthelasma area; The xanthelasma area calculation module is used to calculate the actual area of the xanthelasma area, and specifically includes the following submodules: A contour extraction submodule is used to extract the boundary contour of the xanthelasma area; The connected region analysis submodule is used to determine the connectivity of irregular shapes or overlapping regions using the connected region analysis algorithm; The area calculation submodule is used to calculate the area of the xanthelasma region based on the contour boundary and connected region analysis results.