Tongue picture key feature labeling method based on artificial intelligence

Abstract

The invention relates to the technical field of tongue picture image processing, in particular to a tongue picture key feature labeling method based on artificial intelligence. According to the method, the tongue body region is converted into the LAB space, and the tongue body and tongue coating probability of each pixel point is obtained according to the position feature and the color feature of each pixel point; according to the gradient distribution characteristics and position distribution between each pixel point and the pixel points in the neighborhood range, the tongue crack probability of each pixel point is obtained; according to the color features, the position features and the gradient distribution of the different pixel points, the tongue scab probability of each pixel point is obtained; according to the tongue coating probability, the tongue crack probability and the tongue scab probability of each pixel point, the overall abnormal probability of each pixel point is obtained; and obtaining abnormal pixel points according to the overall abnormal probability of each pixel point, and carrying out feature labeling on the tongue picture. According to the method, the accuracy of key feature labeling is improved by accurately analyzing the abnormal degree of each pixel point.

Description

Technical Field

[0001] The present invention relates to the technical field of tongue image processing, and particularly to an artificial intelligence-based method for annotating key features of tongue images. Background Art

[0002] By observing tongue image features such as the color, thickness, shape, and color of the tongue coating in the tongue image, key bases for disease diagnosis, syndrome differentiation, and curative effect evaluation can be provided. Therefore, it is necessary to annotate the key features.

[0003] In the prior art, joint modeling is performed through three channels in the LAB color space, and tongue body segmentation and key feature classification are synchronously completed with the aid of a deep learning model. However, when relying on a complex network to process multi-channel information, there is computational redundancy, high resource requirements, and single-channel recognition cannot highlight the regional differences between the tongue coating and non-tongue coating areas, resulting in large annotation errors. Summary of the Invention

[0004] In order to solve the technical problems of multi-channel computational redundancy and the inability of single-channel recognition to highlight the regional differences between the tongue coating and non-tongue coating areas, resulting in large annotation errors, the purpose of the present invention is to provide an artificial intelligence-based method for annotating key features of tongue images, and the specific technical solutions adopted are as follows: The present invention proposes an artificial intelligence-based method for annotating key features of tongue images, and the method includes: Obtain a tongue image, where the tongue image includes a tongue body area; Convert the tongue body area to the LAB color space, and obtain the tongue body coating probability of each pixel point according to the position feature and color feature of each pixel point; obtain the tongue body crack probability of each pixel point according to the gradient distribution feature and position distribution between each pixel point and the pixel points within the neighborhood range; obtain the tongue body lesion probability of each pixel point according to the color feature, position feature, and gradient distribution of different pixel points; Obtain the overall abnormality probability of each pixel point according to the tongue body coating probability, tongue body crack probability, and tongue body lesion probability of each pixel point; obtain abnormal pixel points according to the overall abnormality probability of each pixel point, and perform feature annotation on the tongue image.

[0005] Further, the method for obtaining the tongue body coating probability includes: Obtain the position weight of each pixel point according to the position feature of each pixel point; Obtain the B-channel color value corresponding to each pixel point in the tongue body area in the LAB color space, normalize the B-channel color value, and use it as the coating possibility of each pixel point; Weight the coating possibility according to the position weight of each pixel point to obtain the tongue body coating probability of each pixel point.

[0006] Further, the method for obtaining the position weight includes: Taking the direction from the root of the tongue to the tip of the tongue as the vertical direction, the direction of the tongue body from left to right as the horizontal direction, and the midpoint of the bottommost edge of the tongue body as the origin, a tongue body coordinate system is constructed; According to the differences in the coordinate values in each coordinate direction between each pixel point and the origin, the position weight of each pixel point is obtained, and the difference in the coordinate value is negatively correlated with the position weight.

[0007] Further, the method for obtaining the tongue body crack probability includes: Obtaining the gradient amplitude of the corresponding L channel of each pixel point in the LAB color space; normalizing the gradient amplitude of each pixel point as the crack possibility of each pixel point; Obtaining the pixel point with the largest gradient amplitude within the neighborhood range of each pixel point as the reference pixel point, and obtaining a preset number of reference pixel points for each pixel point; According to the crack possibility of each pixel point, the position change characteristics and the gradient direction distribution of all corresponding reference pixel points, the tongue body crack probability of each pixel point is obtained.

[0008] Further, the method for obtaining the tongue body crack probability includes: Obtaining the fluctuation characteristics of the coordinate differences in each coordinate direction between all adjacent reference pixel points corresponding to each pixel point as the local position fluctuation degree in each coordinate direction corresponding to each pixel point; Obtaining the angle fluctuation characteristics of the gradient directions between adjacent reference pixel points as the gradient fluctuation degree; According to the crack possibility of each pixel point, the local position fluctuation degree in all coordinate directions and the gradient fluctuation degree, the tongue body crack probability of each pixel point is obtained. The crack possibility is positively correlated with the tongue body crack probability, and both the local position fluctuation degree and the gradient fluctuation degree are negatively correlated with the tongue body crack probability.

[0009] Further, the method for obtaining the tongue body lesion probability includes: According to the deviation of the A channel color value corresponding to each pixel point in the LAB color space from the A channel color standard value of the standard tongue image, the tongue body color abnormality probability of each pixel point is obtained; According to the distribution of the A channel color values corresponding to all pixel points within the neighborhood range of different pixel points, the growth area of each pixel point is obtained; According to the tongue body color abnormality probability of each pixel point, the gradient fluctuation characteristics of the edge pixel points in the growth area, and the difference in the number of pixel points in the growth area between the axially symmetric pixel points in the tongue body, the tongue body lesion probability of each pixel point is obtained. The tongue body color abnormality probability, the gradient fluctuation characteristics of the edge pixel points in the growth area, and the difference in the number of pixel points are all positively correlated with the tongue body lesion probability.

[0010] Furthermore, the method for obtaining the probability of abnormal tongue body color includes: Obtain the difference between the maximum or minimum value of the A-channel color corresponding to the pixel points in the tongue body area and the standard value of the A-channel color, obtain the difference between the A-channel color value corresponding to each pixel point and the standard value of the A-channel color, and calculate the ratio between the difference result and the maximum difference value as the probability of abnormal tongue body color for each pixel point.

[0011] Furthermore, the method for obtaining the growth area includes: Obtain the difference in the A-channel color values between each pixel point and each pixel point within the neighborhood range, and perform normalization as the degree of similarity. If the degree of similarity is less than the preset similarity threshold, regard the corresponding pixel point as a similar pixel point, obtain all the similar pixel points of each pixel point, and form the growth area.

[0012] Furthermore, the method for obtaining the overall abnormality probability includes: Obtain the product of the tongue coating probability and the tongue crack probability of the tongue body as the first abnormality possibility; obtain the difference between the positive integer 1 and the tongue coating probability of the tongue body, and calculate the product of the difference result and the tongue lesion probability of the tongue body as the second abnormality possibility. Fuse the first abnormality possibility and the second abnormality possibility to obtain the overall abnormality probability of each pixel point.

[0013] Furthermore, the method for obtaining the abnormal pixel points includes: If the overall abnormality probability of each pixel point is greater than the preset abnormality threshold, regard the corresponding pixel point as an abnormal pixel point.

[0014] The present invention has the following beneficial effects: By converting the tongue body area into the LAB color space, the present invention obtains the tongue coating probability of each pixel point according to the position characteristics and color characteristics of each pixel point; obtains the tongue crack probability of each pixel point according to the gradient distribution characteristics and position distribution between each pixel point and the pixel points within the neighborhood range; obtains the tongue lesion probability of each pixel point according to the color characteristics, position characteristics and gradient distribution of different pixel points; obtains the overall abnormality probability of each pixel point according to the tongue coating probability, tongue crack probability and tongue lesion probability of each pixel point; obtains the abnormal pixel points according to the overall abnormality probability of each pixel point, and performs feature annotation on the tongue image. The present invention improves the accuracy of key feature annotation by accurately analyzing the degree of abnormality of each pixel point. Description of the Drawings

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the 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 drawings can be obtained based on these drawings.

[0016] Figure 1 The flowchart of a method for annotating key features of tongue images based on artificial intelligence provided by an embodiment of the present invention.

[0017] Figure 2 The flowchart of a method for obtaining the probability of the tongue body and tongue coating provided by an embodiment of the present invention; Figure 3 The schematic diagram of the tongue body coordinate system provided by an embodiment of the present invention; Figure 4 The flowchart of a method for obtaining the probability of tongue body cracks provided by an embodiment of the present invention; Figure 5 The flowchart of a method for obtaining the probability of tongue body lesions provided by an embodiment of the present invention. Detailed implementation manners

[0018] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following, in combination with the drawings and preferred embodiments, details the specific implementation manners, structures, features and effects of a method for annotating key features of tongue images based on artificial intelligence proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0020] The following specifically describes the specific solution of a method for annotating key features of tongue images based on artificial intelligence provided by the present invention with reference to the drawings.

[0021] Please refer to Figure 1 , which shows the flowchart of a method for annotating key features of tongue images based on artificial intelligence provided by an embodiment of the present invention. The specific method includes: Step S1: Obtain a tongue image, where the tongue image includes a tongue body area.

[0022] In an embodiment of the present invention, in order to provide a basis for tongue diagnosis, it is necessary to observe the tongue coating color, thickness, tongue shape, color and other characteristics of the tongue, and analyze the performance characteristics of the pixel points in the tongue area; first of all, it should be noted that in order to facilitate subsequent image processing, dual polarizers are used to eliminate reflection interference and ensure resolution, so as to obtain an image that truly reflects the details of the tongue coating and tongue quality. At the same time, the image resolution is required to be ≥640×480 to clearly capture tiny features such as cracks and ecchymosis, avoid feature blur and false detection caused by low resolution, and obtain a tongue image, which includes the tongue area.

[0023] It should be noted that, in an embodiment of the present invention, the method for obtaining the tongue area includes: performing semantic segmentation on the tongue body through the MobileNetV3+UNet model to obtain the tongue body area; the specific MobileNetV3+UNet model is a technical means well known to those skilled in the art and will not be elaborated here.

[0024] Step S2: Convert the tongue area to LAB space, and obtain the tongue coating probability of each pixel based on the position characteristics and color characteristics of each pixel; obtain the tongue crack probability of each pixel based on the gradient distribution characteristics and position distribution between each pixel and the pixels in the neighborhood; obtain the tongue spot probability of each pixel based on the color characteristics, position characteristics and gradient distribution of different pixels.

[0025] In order to better obtain the color characteristics therein, the tongue area is converted to the LAB space. It should be noted that the LAB color space represents a more accurate color range than other color spaces. It consists of a brightness channel L and two color channels A and B. A represents the component from green to red. The larger the A channel value, the more red the characteristics. B represents the component from blue to yellow. The larger the B channel value, the more yellow the characteristics.

[0026] The probability of tongue coating appearing in the middle and root of the tongue shows a normal distribution characteristic, that is, the farther away from the root of the tongue, the lower the probability of tongue coating appearing; the tongue coating is usually light yellow or light white, and the non-tongue coating is light red. By analyzing the color channels on the LAB image, the B channel can effectively distinguish the difference between yellow and blue. The larger the value, the more it tends to yellow, thereby quantifying the possibility of the existence of tongue coating; according to the position characteristics and color characteristics of each pixel point, the tongue body and tongue coating probability of each pixel point is obtained.

[0027] Preferably, in one embodiment of the present invention, the method for obtaining the tongue body and tongue coating probability can be found in Figure 2 , which shows a flow chart of a method for obtaining tongue body and tongue coating probability, including: Step S201: Obtain the position weight of each pixel according to the position feature of each pixel.

[0028] Preferably, in an embodiment of the present invention, the method for obtaining the position weight includes: Taking the direction from the root of the tongue to the tip of the tongue as the vertical direction, the direction of the tongue body from left to right as the horizontal direction, and the midpoint of the bottommost edge of the tongue body as the origin, a tongue body coordinate system is constructed, as Figure 3 shown, which shows a schematic diagram of a tongue body coordinate system; According to the difference in the coordinate values in each coordinate direction between each pixel point and the origin, the position weight of each pixel point is obtained, and the difference in the coordinate values is negatively correlated with the position weight.

[0029] In an embodiment of the present invention, by selecting the maximum value of the difference in the coordinate values between the pixel points in the vertical direction and the origin, and the maximum value of the difference in the coordinate values between the pixel points in the horizontal direction, calculating the ratio of the difference in the coordinate values in each coordinate direction between each pixel point and the origin to the corresponding maximum value of the difference, and performing a negative correlation mapping, multiplying the negative correlation mapping results in the vertical direction and the horizontal direction, the position weight of each pixel point is obtained; Therefore, the correlation relationship between the difference in the coordinate values in the coordinate direction and the position weight is formed, so that the greater the difference in the coordinate values in the coordinate direction, the closer to the edge of the tongue body, the farther from the tongue coating area, and the smaller the corresponding position weight.

[0030] Step S202: Obtain the B-channel color value corresponding to each pixel point in the tongue body area in the LAB color space, and normalize the B-channel color value as the tongue coating possibility of each pixel point.

[0031] In an embodiment of the present invention, by calculating the difference between the B-channel color value of each pixel point and the minimum value, and taking the ratio of the difference result to the extreme difference of the B-channel color value, that is, normalizing the B-channel color value of each pixel point, the greater the B-channel color value, the more likely it is to exhibit the yellow feature, the more likely there is a tongue coating, and the greater the tongue coating possibility.

[0032] Step S203: Weight the tongue coating possibility according to the position weight of each pixel point to obtain the tongue body tongue coating probability of each pixel point.

[0033] In an embodiment of the present invention, calculate the product of the tongue coating possibility and the position weight to obtain the tongue body tongue coating probability. Therefore, the correlation relationship between the tongue coating possibility, the position weight, and the tongue body tongue coating probability is formed. The greater the tongue coating possibility, the greater the position weight, and the greater the tongue body tongue coating probability.

[0034] The pixel brightness at the cracks of the tongue coating will show significant changes, manifested as higher gradients; and the pixel points on the cracks are arranged in a linear straight line, and the higher the consistency of the gradient distribution direction; therefore, according to the gradient distribution characteristics and position distribution between each pixel point and the pixel points within the neighborhood range, the tongue crack probability of each pixel point is obtained.

[0035] Preferably, in an embodiment of the present invention, for the method of obtaining the tongue crack probability, please refer to Figure 4 , which shows a flowchart of a method for obtaining the tongue crack probability, including: Step S401: Obtain the gradient amplitude of the L channel corresponding to each pixel point in the LAB color space; normalize the gradient amplitude of each pixel point as the crack possibility of each pixel point.

[0036] In the L channel, the brightness at the pixel points at the cracks of the tongue coating has changed significantly. The larger the gradient amplitude, the more likely it is a crack.

[0037] It should be noted that, in some embodiments of the present invention, the horizontal gradient and vertical gradient of each pixel point can be obtained through the Sobel operator or the Prewitt operator, and then the corresponding gradient amplitude and direction can be obtained; the specific means are well-known technical means to those skilled in the art and will not be elaborated here.

[0038] It should be noted that, in the embodiments of the present invention, normalization can be performed through maximum-minimum normalization or sigmoid mapping function, etc. The specific means are well-known technical means to those skilled in the art and will not be elaborated here.

[0039] Step S402: Obtain the pixel point with the largest crack possibility within the neighborhood range of each pixel point as the reference pixel point, and obtain a preset number of reference pixel points for each pixel point.

[0040] The pixel points with greater crack possibility are more likely to show crack phenomena. Selecting adjacent and continuous pixel points for analysis helps to analyze whether they exhibit crack characteristics.

[0041] It should be noted that, in an embodiment of the present invention, the neighborhood range is a range composed of the pixel points within the 8-neighborhood of each pixel point with each pixel point as the reference, excluding the reference pixel point; the size of the preset number is 8; in other embodiments of the present invention, the neighborhood range and the size of the preset number can be specifically set according to specific situations and will not be limited and elaborated here.

[0042] Step S403: Obtain the crack probability of each pixel point according to the crack possibility of each pixel point, the position change characteristics of all corresponding reference pixel points, and the gradient direction distribution.

[0043] Preferably, in an embodiment of the present invention, the method for obtaining the crack probability includes: Obtain the fluctuation characteristics of the coordinate differences between each pixel point and all adjacent reference pixel points in each coordinate direction as the local position fluctuation degree of each pixel point in each coordinate direction; Obtain the angle fluctuation characteristics of the gradient directions between adjacent reference pixel points as the gradient fluctuation degree; According to the crack possibility of each pixel point, the local position fluctuation degrees in all coordinate directions, and the gradient fluctuation degree, obtain the tongue crack probability of each pixel point. The crack possibility is positively correlated with the tongue crack probability, and both the local position fluctuation degree and the gradient fluctuation degree are negatively correlated with the tongue crack probability.

[0044] It should be noted that in an embodiment of the present invention, the fluctuation characteristics can be represented by calculating the standard deviation. The larger the standard deviation, the larger the fluctuation characteristics; the smaller the standard deviation, the smaller the fluctuation characteristics. In other embodiments of the present invention, the fluctuation characteristics can also be represented by calculating the variance or range. The specific means are well-known technical means to those skilled in the art and will not be elaborated here.

[0045] Considering that the gradient direction and distribution of the pixel points on the crack tend to be straight lines, and the corresponding position change and gradient change are smaller. In an embodiment of the present invention, a negative correlation mapping is performed on each fluctuation degree, and the sum of the negative correlation mapping results in all coordinate directions is obtained as the first cumulative value; calculate the product of the first cumulative value, the gradient fluctuation degree, and the crack possibility as the tongue crack probability of each pixel point; Therefore, the correlation relationship between the crack possibility, the local position fluctuation degrees in all coordinate directions, the gradient fluctuation degree, and the tongue crack probability is formed. The greater the crack possibility, the smaller the local position fluctuation degree, the smaller the gradient fluctuation degree, and the greater the tongue crack probability.

[0046] The non-tongue coating area presents a red feature. The physiological red area of the normal tongue image is bilaterally symmetric along the tongue's central axis, and there is no significant difference in color difference between the two sides. However, if there are pathological plaques, the red area shows a random distribution feature and does not have a symmetry law. Therefore, the tongue lesion probability of the pixel points is quantified based on the color distribution features, position features, and gradient directions of different pixel points. According to the color distribution features, position features, and gradient directions of different pixel points, the tongue lesion probability of each pixel point is obtained.

[0047] Preferably, in an embodiment of the present invention, for the method of obtaining the tongue lesion probability, please refer to Figure 5 , which shows a flowchart of a method for obtaining the tongue lesion probability, including: Step S501: Obtain the tongue color abnormality probability of each pixel according to the deviation of the A-channel color value corresponding to each pixel in the LAB color space from the A-channel color standard value of the standard tongue image.

[0048] Preferably, in an embodiment of the present invention, the method for obtaining the tongue color abnormality probability includes: Obtain the difference between the maximum or minimum value of the A-channel color corresponding to the pixel in the tongue region and the A-channel color standard value, obtain the difference between the A-channel color value corresponding to each pixel and the A-channel color standard value, and calculate the ratio between the difference result and the maximum difference value as the tongue color abnormality probability of each pixel.

[0049] Among them, the difference between the maximum or minimum value of the A-channel color corresponding to the pixel in the tongue region and the A-channel color standard value can reflect the degree of the largest deviation of the A-channel color value of the tongue region from the standard tongue image, and is used to quantify the color deviation of each pixel. The greater the deviation from the standard color, the greater the tongue color abnormality probability.

[0050] It should be noted that in the embodiment of the present invention, the A-channel color standard value of the standard tongue image can be pre-obtained by the implementer according to relevant clinical data.

[0051] Step S502: Obtain the growth region of each pixel according to the distribution of the A-channel color values corresponding to all pixels within the neighborhood range of different pixels.

[0052] Preferably, in an embodiment of the present invention, the method for obtaining the growth region includes: Obtain the difference between the A-channel color values between each pixel and each pixel within the neighborhood range, and perform normalization as the similarity degree; If the similarity degree is less than the preset similarity threshold, the corresponding pixel is regarded as a similar pixel, and all similar pixels of each pixel are obtained to form the growth region.

[0053] It should be noted that in an embodiment of the present invention, the size of the preset similarity threshold is 0.3; in other embodiments of the present invention, the size of the preset similarity threshold can be specifically set according to specific situations, and will not be limited and elaborated here.

[0054] Step S503: Obtain the tongue lesion probability of each pixel according to the tongue color abnormality probability of each pixel, the gradient fluctuation characteristics of the edge pixels in the growth region, and the difference in the number of pixels in the growth region between the axially symmetric pixels in the tongue. The tongue color abnormality probability, the gradient fluctuation characteristics of the edge pixels in the growth region, and the pixel number difference are all positively correlated with the tongue lesion probability.

[0055] In an embodiment of the present invention, the product of the probability of abnormal tongue body color, the gradient fluctuation feature of the edge pixel points in the growth region, and the difference in the number of pixel points is used as the probability of tongue body lesion for each pixel point. Therefore, the correlation relationship between the probability of abnormal tongue body color, the gradient fluctuation feature of the edge pixel points in the growth region, the difference in the number of pixel points, and the red spots is formed, that is, the greater the red probability, the greater the gradient fluctuation feature of the edge pixel points in the growth region, the more likely to have an obvious boundary, the greater the difference in color distribution, the greater the difference in the number of pixel points in the growth region between symmetric pixel points, indicating that the growth region between symmetric pixel points is asymmetric, the greater the possibility of random distribution, and the more likely there are pathological plaques.

[0056] Step S3: Obtain the overall abnormality probability of each pixel point according to the probability of tongue coating on the tongue body, the probability of tongue cracks, and the probability of tongue body lesions for each pixel point; obtain the abnormal pixel points according to the overall abnormality probability of each pixel point, and perform feature annotation on the tongue image.

[0057] The higher the probability of tongue coating on the tongue body, the relatively higher the probability of cracks, which makes the red manifestation be covered by the tongue coating, and the smaller the characteristics of the red manifestation. Therefore, the overall abnormality probability of each pixel point is obtained according to the probability of tongue coating on the tongue body, the probability of tongue cracks, and the probability of tongue body lesions for each pixel point.

[0058] Preferably, in an embodiment of the present invention, the method for obtaining the overall abnormality probability includes: Obtain the product of the probability of tongue coating on the tongue body and the probability of tongue cracks as the first abnormality possibility; obtain the difference between the positive integer 1 and the probability of tongue coating on the tongue body, and calculate the product between the difference result and the probability of tongue body lesions as the second abnormality possibility. Based on this, the greater the probability of tongue coating on the tongue body, the higher the probability of cracks, the greater the credibility of the probability of tongue cracks, and the more it is adjusted upwards. The greater the probability of tongue coating on the tongue body, the more likely it is to cover the physiological red area, the smaller the credibility of the probability of tongue body lesions, and the smaller the characteristics shown, and it is adjusted downwards.

[0059] Fuse the first abnormality possibility and the second abnormality possibility to obtain the overall abnormality probability of each pixel point.

[0060] It should be noted that in some embodiments of the present invention, the first abnormality possibility and the second abnormality possibility can be fused by addition or multiplication methods to form a relationship where the greater the first abnormality possibility, the greater the second abnormality possibility, and the greater the overall abnormality probability; the specific means are well-known technical means to those skilled in the art and will not be elaborated here.

[0061] The overall abnormality probability more comprehensively and accurately reflects the abnormal situation at each pixel point. The greater the overall abnormality probability, the more obvious the tongue coating crack or red spot characteristics are, and the more it can reflect the state of the tongue body.

[0062] Preferably, in an embodiment of the present invention, the method for obtaining abnormal pixel points includes: If the overall abnormal probability of each pixel point is greater than a preset abnormal threshold, the corresponding pixel point is regarded as an abnormal pixel point.

[0063] It should be noted that, in an embodiment of the present invention, the size of the preset abnormal threshold is 0.8; in other embodiments of the present invention, the size of the preset abnormal threshold can be specifically set according to specific situations, and no limitation and elaboration are made here.

[0064] Based on this, by obtaining the abnormal pixel points on the tongue body area, the more the abnormal characteristics of the tongue body area can be reflected, and by labeling the abnormal pixel points, the abnormal labeling of the tongue image is realized, which helps to provide a key basis for the subsequent diagnosis of the tongue body.

[0065] In summary, the present invention converts the tongue body area into the LAB color space, obtains the tongue coating probability of each pixel point according to the position characteristics and color characteristics of each pixel point; obtains the tongue crack probability of each pixel point according to the gradient distribution characteristics and position distribution between each pixel point and the pixel points within the neighborhood range; obtains the tongue lesion probability of each pixel point according to the color characteristics, position characteristics and gradient distribution of different pixel points; obtains the overall abnormal probability of each pixel point according to the tongue coating probability, tongue crack probability and tongue lesion probability of each pixel point; obtains the abnormal pixel points according to the overall abnormal probability of each pixel point, and performs feature labeling on the tongue image. The present invention improves the accuracy of key feature labeling by accurately analyzing the abnormal degree of each pixel point.

[0066] It should be noted that: the above sequence of embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0067] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. An artificial intelligence-based key feature annotation method for tongue images, characterized in that, The method includes: Obtaining a tongue image, which includes a tongue body region; Converting the tongue body region to the LAB color space. According to the position features and color features of each pixel, the tongue coating probability of each pixel is obtained. According to the gradient distribution features and position distribution between each pixel and the pixels within its neighborhood range, the tongue crack probability of each pixel is obtained. According to the color features, position features, and gradient distribution of different pixels, the tongue lesion probability of each pixel is obtained; According to the tongue coating probability, tongue crack probability, and tongue lesion probability of each pixel, the overall abnormality probability of each pixel is obtained. According to the overall abnormality probability of each pixel, abnormal pixels are obtained, and the tongue image is feature-labeled.

2. The method for annotating key features of tongue images based on artificial intelligence according to claim 1, wherein The method for obtaining the tongue coating probability of the tongue body includes: According to the position features of each pixel, the position weight of each pixel is obtained; Obtaining the B-channel color value corresponding to each pixel in the tongue body region in the LAB color space, normalizing the B-channel color value, and using it as the coating possibility of each pixel; Weighting the coating possibility according to the position weight of each pixel to obtain the tongue coating probability of each pixel.

3. The method for annotating key features of tongue images based on artificial intelligence according to claim 2, characterized in that, The method for obtaining the position weight includes: Taking the direction from the root of the tongue to the tip of the tongue as the vertical direction, the direction of the tongue body from left to right as the horizontal direction, and the midpoint of the bottom edge of the tongue body as the origin, a tongue body coordinate system is constructed; According to the differences in the coordinate values of each pixel and the origin in each coordinate direction, the position weight of each pixel is obtained, and the difference in the coordinate value is negatively correlated with the position weight.

4. A method for annotating key features of tongue images based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the tongue crack probability of the tongue body includes: Obtaining the gradient magnitude of the L-channel corresponding to each pixel in the LAB color space; normalizing the gradient magnitude of each pixel and using it as the crack possibility of each pixel; Obtaining the pixel with the largest gradient magnitude within the neighborhood range of each pixel as the reference pixel, and obtaining a preset number of reference pixels for each pixel; According to the crack possibility of each pixel, the position change features and gradient direction distribution corresponding to all reference pixels, the tongue crack probability of each pixel is obtained.

5. The method for annotating key features of tongue images based on artificial intelligence according to claim 4, wherein, The method for obtaining the tongue crack probability of the tongue body includes: Obtaining the fluctuation feature of the coordinate difference between each pixel and all adjacent reference pixels in each coordinate direction as the local position fluctuation degree of each pixel in each coordinate direction; Obtaining the angle fluctuation feature of the gradient direction between adjacent reference pixels as the gradient fluctuation degree; According to the crack possibility of each pixel, the local position fluctuation degree and gradient fluctuation degree corresponding to all coordinate directions, the tongue crack probability of each pixel is obtained. The crack possibility is positively correlated with the tongue crack probability, and both the local position fluctuation degree and the gradient fluctuation degree are negatively correlated with the tongue crack probability.

6. A method for annotating key features of tongue images based on artificial intelligence according to claim 1, characterized in that The method for obtaining the tongue lesion probability of the tongue body includes: According to the deviation of the A-channel color value corresponding to each pixel in the LAB color space from the A-channel color standard value of the standard tongue image, the tongue body color abnormality probability of each pixel is obtained; Obtain the growth region of each pixel according to the distribution of the A-channel color values corresponding to all pixels within the neighborhood range of different pixels. Obtain the probability of tongue lesion for each pixel according to the probability of abnormal tongue color of each pixel, the gradient fluctuation characteristics of the edge pixels in the growth region, and the difference in the number of pixels in the growth region between the axially symmetric pixels in the tongue. The probability of abnormal tongue color, the gradient fluctuation characteristics of the edge pixels in the growth region, and the difference in the number of pixels are all positively correlated with the probability of tongue lesion.

7. A method for annotating key features of tongue images based on artificial intelligence according to claim 6, characterized in that, The method for obtaining the probability of abnormal tongue color includes: Obtain the difference between the maximum or minimum value of the A-channel color corresponding to the pixels in the tongue region and the standard value of the A-channel color, obtain the difference between the A-channel color value corresponding to each pixel and the standard value of the A-channel color, and calculate the ratio between the difference result and the maximum difference value as the probability of abnormal tongue color for each pixel.

8. A method for annotating key features of tongue images based on artificial intelligence according to claim 6, characterized in that, The method for obtaining the growth region includes: Obtain the difference in the A-channel color values between each pixel and each pixel within the neighborhood range, and perform normalization as the degree of similarity. If the degree of similarity is less than the preset similarity threshold, regard the corresponding pixel as a similar pixel, and obtain all similar pixels of each pixel to form the growth region.

9. A method for annotating key features of tongue images based on artificial intelligence according to claim 1, characterized in that The method for obtaining the overall abnormality probability includes: Obtain the product of the tongue coating probability and the tongue crack probability as the first abnormality possibility; obtain the difference between the positive integer 1 and the tongue coating probability, and calculate the product of the difference result and the tongue lesion probability as the second abnormality possibility. Fuse the first abnormality possibility and the second abnormality possibility to obtain the overall abnormality probability of each pixel.

10. A method for annotating key features of tongue images based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the abnormal pixel includes: If the overall abnormality probability of each pixel is greater than the preset abnormality threshold, regard the corresponding pixel as an abnormal pixel.

Images