Ear image acquisition device and ear image recognition analysis method thereof

Abstract

The application discloses an ear image collecting device and an ear image recognition and analysis method thereof, and comprises a handle and a cover body, wherein the cover body is connected with the end of the handle; the cover body is in the shape of a horn and is sequentially provided with a light reflection part, a folding part and a light shielding part from inside to outside, and a standard color ring is arranged on the junction of the front end of the folding part and the light shielding part along the circumference. The beneficial effect is that the collecting device can realize light shielding and light supplement when collecting the ear image, ensure the stability of the light source, reduce the hue deviation, adjust the angle between the cover body and the ear through the folding and telescopic design of the folding part, obtain the image perpendicular to the ear, and make the ear image clear and complete. Through the ear image recognition and analysis method, the effective recognition, segmentation and feature point matching of the ear region image can be improved, 91 feature points of the ear image are obtained, the ear region is recognized and trained, and the recognition accuracy of the ear image acupoint positioning is higher than that of the traditional method.

Description

TECHNICAL FIELD

[0001] The present application relates to the ear image collection and analysis field, in particular to an ear image collection device and an ear image recognition and analysis method thereof. BACKGROUND

[0002] Ear observation has very important significance in ear acupuncture diagnosis, which is a method of diagnosing diseases by observing the change rules of some characteristics of normal and abnormal ear. The observation content includes the overall diagnosis of ear appearance size, shape, structure, position, thickness, color, dryness and wetness, desquamation, protuberance, nodule, papule, and blood vessel filling degree, and then the specific diagnosis is carried out through the observation of the specific change rules of each ear acupuncture part and positioning. Ear acupuncture observation is a characteristic technology of ear acupuncture diagnosis, which is accurate, simple and easy to master, and is an effective means for us to quickly find the potential disease risk of human body.

[0003] However, the ear acupuncture observation technology is difficult to master, and a large number of clinical practice observation cases are needed, which is very difficult for beginners. With the rapid development of computer artificial intelligence image processing technology, a new ear diagnosis method, ear image diagnosis method, has been developed. Among them, ear image recognition involves ear image positioning and region segmentation, which is a key link of intelligent traditional Chinese medicine ear acupuncture diagnosis and treatment. It mainly uses a specially designed ear image collection device to obtain high-definition ear image, and then manually marks different ear image features, and then uses cloud computing and artificial intelligence image learning algorithm to automatically recognize the features and morphological changes of acupuncture point area, color difference, protrusion hyperplasia, and blood vessel dilation, and mark the features of the abnormal area, so as to provide auxiliary diagnosis reference for clinicians.

[0004] However, the traditional ear acupuncture observation method mainly relies on the visual observation of doctors for judgment and analysis, and the diagnosis result is limited by the knowledge level and diagnosis technology of doctors, and is affected by the external environment such as light and temperature at that time. In addition, the current ear acupuncture observation lacks objective evaluation basis, and the description of characteristics is rough, lacks objective quantification, is difficult to form a stable standard, and is not easy to understand and teach, which seriously restricts the application and development of ear acupuncture observation. SUMMARY

[0005] The purpose of the present application is to provide an ear image collection device and an ear image recognition and analysis method thereof to solve the above problems.

[0006] The present application achieves the above-mentioned purpose through the following technical solutions:

[0007] An ear image collection device, comprising a handle and a cover body, wherein the cover body is connected with the end of the handle.

[0008] The cover body is trumpet-shaped and sequentially comprises a light reflection part, a folding part and a light-proof part from inside to outside, and the standard color ring is arranged along the circumference at the junction of the front end of the folding part and the light-proof part;

[0009] The handle is provided with a control switch, an internal control circuit board, a lens module mounted at the end of the handle, the lens module being electrically connected with the control circuit board, an annular flexible light source being arranged at the periphery of the lens module, and a soft light cover being arranged on the lamp bead surface of the annular flexible light source.

[0010] Preferably, an annular scale card is arranged on the light-proof part adjacent to the standard color ring, and the standard color ring and the scale card are connected with the inner wall of the light-proof part by adhesion.

[0011] Preferably, the cover body is integrally made of soft rubber, and the inner side surfaces of the light reflection part, the folding part and the light-proof part are coated with white light reflection coating.

[0012] Preferably, the edges of the light-proof part are provided with two convex edges and two concave edges, the two convex edges corresponding to each other, the two concave edges corresponding to each other, the convex edges and the concave edges being staggered and connected by smooth curves.

[0013] Preferably, the radial section of the folding part is wavy, capable of being bent, stretched and folded, and the folding angle is 3-5°.

[0014] Preferably, the control circuit board is provided with a wireless connection component and a wired connection component, and the wireless connection component and the wired connection component can be wirelessly or wiredly connected with a computer.

[0015] The back of the handle is further provided with a display electrically connected with the control circuit board, capable of displaying the collected ear image information in real time.

[0016] Preferably, the cover body and the handle are detachably connected by threads or buckles.

[0017] Preferably, the lens module is a short-focus lens, adopts F2.0 large aperture, large-angle lens, and 1300W pixel CMOS active pixel type superimposed image sensor and square pixel array, adopts EXMOR RS, realizes back light illumination imaging pixel structure, can realize A / D converter circuit and high-sensitivity low-noise image high-speed image acquisition, and further adopts spatial multiplexing exposure, can realize high dynamic range still image acquisition.

[0018] The filter adopts R, G and B pigment primary color mosaic filter.

[0019] Using the above-described ear image acquisition device, a trumpet-shaped cover is placed over the ear. The angle of the cover is adjusted by the folding part, ensuring that the cover faces the auricle directly without obstructing the inner edge of the helix, thus covering the entire auricle. By controlling the lens module and a stable ring-shaped flexible light source through a control switch, the device achieves consistent light source in the acquired images, resulting in clear images of the auricle surface structure. During acquisition, the cover mitigates the influence of external optical fibers on the shooting environment, while the internal flexible supplementary lighting design resolves highlights and shadows on the protruding and recessed areas of the auricle during photography. Furthermore, a standard colorimetric card ring with an 18% gray card layer is designed at the front of the inner layer of the cover, allowing for simultaneous verification of the auricle color value during photography, ensuring that the color difference in the acquired ear image is controlled within a minimal deviation range.

[0020] The method for acquiring ear images is as follows:

[0021] (1) The upper edge of the ear tip is parallel to the hair bun;

[0022] (2) Center the lens on the helix and conchae cymbidium, take the front view of the auricle, and leave 1-2 cm around the auricle image size;

[0023] (3) The vertical range of the image is the distance between the highest point of the upper edge of the auricle and the lowest point of the earlobe;

[0024] (4) The upper part of the horizontal width of the photograph is the part where the upper edge of the auricle attaches to the side of the head and the most prominent part of the posterior edge of the auricle;

[0025] (5) The lower horizontal width of the photograph is the distance between the lower edge of the earlobe attaching to the cheek and the back edge of the earlobe;

[0026] (6) Adjust the folding part of the cover so that the shooting angle and the tilt of the head and face are consistent with the angle of the ear and skull, so as not to block the inner edge of the antihelix.

[0027] A method for recognizing and analyzing ear images acquired by an ear image acquisition device, comprising the following steps:

[0028] Step 1: The ear image photographs obtained by the standardized capture device are segmented into the outer contour of the auricle and the face, and the tilt angle of the auricle is corrected according to the auricle marker line to obtain the vector of the initial ear shape of the ear image.

[0029] Step 2: Establish a sample library of auricular acupoint regions, physiological features of the auricle, contour points, and pathological features of auricular acupoints. Based on labeled sample ear images and machine learning training, achieve automatic identification and positioning of auricular acupoint regions.

[0030] Step 3: Using convolution algorithm, local grayscale model, and minimum variance comparison, the image algorithm is optimized to complete the extraction of ear image features, including color features, texture features, and morphological features. The images of the feature areas are then automatically segmented and saved separately.

[0031] Step four: according to the analysis results of extracting ear image features, the features of the auricular point area are compared with the feature database to form a pre-diagnosis analysis report.

[0032] Further, the step two of establishing auricular point partition, ear physiological feature and ear point recognition training includes the following steps:

[0033] S1: identify the physiological feature points of the ear;

[0034] S2: combine the edge segmentation of auricular point national standard partition to obtain the corresponding 91 ear landmark points, and draw lines between the landmark points to divide different auricular point partitions;

[0035] S3: train and recognize the labeled data to establish a Shape model, and use M pictures as training samples, each picture is labeled with N (91) feature point coordinates expressed as (x, y), and the expression form of each sample shape vector is defined as: Each sample picture has a set of shape vectors S, and the length is 2*N, i represents the i-th sample;

[0036] According to the diagonal line length of 200 of the picture, the overall size of the picture is scaled, the length of the picture is kept unchanged, and finally all the pictures have consistent diagonal line length, and the feature vectors are also transformed in the same way;

[0037] After the picture is preprocessed, the active shape model is trained, and the generation process is to first execute GPA, and then perform PCA to obtain the active shape model function expressed as: ;

[0038] Wherein, Savg is the average shape of the entire model, pi is a weight indicating how much change, and Si is the shape change indicating the overall direction of movement of all feature points;

[0039] Further, first, perform PGA to align the data; first, set two samples, the first is the template sample, and the second is the sample to be aligned, and the following steps are completed to align all samples to the template sample;

[0040] Delete the data translation component, for each group of samples Calculate the mean value of the position of all feature points , then subtract the mean value of the overall feature point set to remove the translation component of the data;

[0041] Delete the scaling component of the data, calculate the standard deviation of the feature points in each group of samples, and divide the overall data by the standard deviation to obtain the data without scaling;

[0042] Delete the rotation component, find a rotation angle that is the least biased of two samples, first output the covariance matrix of the data, and then SVD decomposition, R=UV^T; Where, R is the rotation matrix executed, multiplied by the rotation matrix can rotate the data to the angle of the template;

[0043] Secondly, PCA principal component analysis is carried out, first, the center of all the aligned features is centralized, and the function is expressed as: Wherein, M is the total amount of quantity, Indicates the i-th feature of the j-th sample;

[0044] The covariance matrix is generated, which shows the relationship between different feature points. If the value of the corresponding position is large, it proves that the coincidence degree of the change of two features is high, and the similarity is strong;

[0045]

[0046] Then, eigenvalue decomposition is carried out, and the eigenvectors corresponding to the largest Ls characteristic values (S1, S2,..., Sls') are taken out. Each eigenvector is 182-dimensional, indicating the overall change direction of the model, so the shape model is composed of: ;

[0047] Wherein, Savg is the model obtained by averaging each aligned feature, which is also the average model, Si represents each change, and pi represents how much each change affects.

[0048] The beneficial effects are that the light can be avoided and the light can be supplemented when the ear image is collected by the collecting device, the light source is stable, the hue deviation is reduced, the angle between the cover body and the ear can be adjusted through the folding design of the folding part, the image perpendicular to the ear is obtained, and the ear image is clear and complete.

[0049] Through the ear image recognition analysis method, the effective recognition, segmentation and feature point matching of the ear acupoint region image can be improved, 91 feature points of the ear image are obtained, so that the ear acupoint region is recognized and trained, the recognition accuracy of the ear acupoint positioning is higher than that of the traditional method, and the feature recognition more embodies the relevance with diseases and symptoms, so that more accurate analysis is obtained. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0051] Figure 1is the stereoscopic structure diagram of the ear image acquisition device of the present application;

[0052] Figure 2 is the front view structure diagram of the ear image acquisition device of the present application;

[0053] Figure 3 is the left view structure diagram of the ear image acquisition device of the present application;

[0054] Figure 4 is the rear view structure diagram of the ear image acquisition device of the present application with a display;

[0055] Figure 5 is the scale card diagram of the ear image acquisition device of the present application;

[0056] Figure 6 is the ear lobe and face part segmentation diagram in the recognition analysis method of the present application;

[0057] Figure 7 is the ear acupoint partition diagram in the recognition analysis method of the present application;

[0058] Figure 8 is the ear lobe landmark point diagram in the recognition analysis method of the present application;

[0059] Figure 9 is the comparison diagram before and after model training in the recognition analysis method of the present application;

[0060] Figure 10 is the ear image feature extraction diagram in the recognition analysis method of the present application;

[0061] Figure 11 is the flow chart of the recognition analysis method of the present application.

[0062] The following is the explanation of the reference signs:

[0063] 1, handle; 2, control switch; 3, cover body; 3a, light reflection part; 3b, folding part; 3c, light shielding part; 4, standard color ring; 5, annular flexible light source; 6, lens module; 7, scale card; 8, display. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical scheme and advantages of the present application more clear, the technical scheme of the present application will be described in detail below. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor, belong to the scope protected by the present application.

[0065] Reference is made to Figures 1-5As shown, the present application provides an ear image acquisition device, which comprises a handle 1 and a cover body 3, the cover body 3 is connected with the end of the handle 1, and the cover body 3 is detachably connected with the handle 1 through threads or buckles. The cover body 3 is trumpet-shaped and sequentially provided with a reflecting part 3a, a folding part 3b and a light-proof part 3c from inside to outside, the cover body 3 is integrally made of soft rubber material, and the inner side surfaces of the reflecting part 3a, the folding part 3b and the light-proof part 3c are coated with white reflecting coating, so that the light is more uniform.

[0066] The longitudinal section of the folding part 3b is wavy, which can be folded by bending and stretching, and the folding angle is 3-5°, which is convenient for bending at various angles. The angle between the auricle and the face of each person will form a certain angle, that is, the auricranial angle, and the angle of the auricranial angle of most people is in the range of 15-80 degrees (divided into three types, less than 30 degrees is close type, 30-60 degrees is normal type, and more than 60 degrees is flared type), and when folding, the angle of bending can be roughly calculated according to the number of folding layers, so that the shooting angle is more easily grasped. Therefore, in order to obtain the image perpendicular to the auricle, it is necessary to adjust the shooting angle of the camera according to the auricranial angle of the auricle. The cover body 3 in the present application has a folding part 3b in the middle, which can be folded and stretched, and adjusts different special positions in front of the ear and behind the ear, which is convenient for the shooting of the lens module 6.

[0067] The standard color card ring 4 is provided at the junction of the front end of the folding part 3b and the light-proof part 3c along the circumference, the standard color card ring 4 is pasted at the junction of the folding part 3b and the light-proof part 3c, and an 18% gray card layer is adopted. The handle 1 is provided with a control switch 2, and an internal control circuit board is arranged, the control switch 2 can control the circuit to realize four-gear brightness adjustment function (high, medium, low and off), and the light source brightness can be manually adjusted according to different collection environments. The end of the handle 1 is provided with a lens module 6, the lens module 6 is electrically connected with the control circuit board, the periphery of the lens module 6 is provided with a ring-shaped flexible light source 5, and the lamp bead surface of the ring-shaped flexible light source 5 is provided with a soft light cover. The soft light cover can fade the bright highlights, reduce the reflection caused by direct light, and combine the light-proof and reflection technologies in the design of the cover body 3, so that the light is uniform and soft while blocking the external light source. The internal light source is also refracted through the design of the cover body 3, and the inner white reflecting coating acts as a double-acting reflecting plate, so that the light is uniform and soft. The light entering the concave part of the auricle is ensured, the shadow is reduced, the light scattering of the convex part is ensured, and the contrast of the highlights and the shadow is reduced.

[0068] Further, as shown in FIG. 6, the present application also provides an ear image acquisition device, which comprises a handle 1 and a cover body 3, the cover body 3 is connected with the end of the handle 1, and the cover body 3 is detachably connected with the handle 1 through threads or buckles. Figure 4As shown, the light-avoiding part 3c is provided with an annular scale card 7 next to the standard color card 4 ring, and the standard color card ring 4 and the scale card 7 are connected with the inner wall of the light-avoiding part 3c by adhesive. The scale card 7 is used as a size reference point, so that after the ear image is taken, the total length of the ear, the upper width of the ear, the upper width of the earlobe, and other data can be calculated on the computer, for statistical analysis (the scale card is like Figure 5 As shown, the scale line is marked, and the dashed line actually marks the inner wall of the annular card.

[0069] Further, in order to improve the sealing performance of the cover body 3 in contact with the edge of the ear, the edge of the light-avoiding part is provided with two convex edges and two concave edges, the two convex edges correspond to each other, the two concave edges correspond to each other, the convex edges and the concave edges are staggered and connected by a smooth curve. The convex edge is slightly higher than the concave edge. When adjusting the shooting angle, the concave edge is the bending direction of the cover body, and the sealing performance is better.

[0070] The control circuit board is provided with a wireless connection component and a wired connection component, which can be connected to the computer in a wireless or wired manner, and the collected ear image information and data information can be transmitted to the computer system in real time for viewing on the computer. At the same time, the back of the handle 1 is also provided with a display 8 (as shown in Figure 4 The display 8 is electrically connected with the control circuit board and can display the collected ear image information in real time. The collector can observe the ear image on the display in real time and adjust the operation according to the clarity of the ear image, the completeness of the collection, and the color difference.

[0071] As a preferred embodiment, the lens module 6 is a short-focus lens with a large aperture of F2.0, a large-angle lens, and a 1300W pixel CMOS active pixel type superimposed image sensor and a square pixel array. EXMOR RS is adopted through a backlit imaging pixel structure, which can realize A / D converter circuit and high-sensitivity low-noise image high-speed image acquisition. Spatial multiplexing exposure is also adopted to realize high dynamic range still image acquisition. The filter adopts an R, G, B pigment primary color mosaic filter.

[0072] In clinical ear image collection, the picture effect is often affected by the environment and light, and the sharpness of the lens is particularly important. Compared with the traditional CMOS image sensor, the lens module 6 in the present application can collect images at a high speed, support phase focusing, and has a fastest speed of 0.2s, three power supply voltages, analog 2.7V, digital 1.2V and 1.8V input / output interfaces, and realizes low power consumption, stable performance, and leads in dark light capability, power consumption and shooting speed.

[0073] The above ear image collection device, the horn-shaped cover 3 covers the ear, the angle of the cover 3 is adjusted through the folding part 3b, the cover 3 is opposite to the auricle, and the inner edge of the helix is not blocked, and the whole auricle is covered inside. By controlling the switch 2 to control the lens module 6 and the stable annular flexible light source 5, the image obtained by the device is consistent, and the surface structure of the auricle is clear. When collecting, the influence of external optical fiber on the shooting environment is solved by the cover 3, and the highlight and shadow of the protruding and recessed parts of the auricle during shooting are solved by the internal flexible light supplement design; in addition, a standard color card ring 4 is designed at the position of the folding part 3b in the inner layer of the cover 3, the standard color card ring 4 adopts 18% gray card layer, which can verify the color value of the auricle during shooting, and ensure that the color difference of the collected ear image is controlled within the minimum deviation range.

[0074] The method for collecting the ear image is as follows:

[0075] (1) The upper edge of the ear tip is parallel to the hair bun;

[0076] (2) The lens is centered on the helix foot and the concha crus, and the front of the auricle is taken, and the size of the auricle image is left out by 1-2 cm;

[0077] (3) The upper and lower height range of shooting is the distance between the highest point of the auricle upper edge and the lowest point of the ear lobe;

[0078] (4) The upper part of the horizontal width range of shooting is the distance between the auricle upper edge attached to the head side and the most protruding part of the auricle rear edge;

[0079] (5) The lower part of the horizontal width range of shooting is the distance between the ear lobe lower edge attached to the cheek and the ear lobe rear edge;

[0080] (6) Adjust the folding part of the cover, so that the shooting angle is consistent with the inclination of the head and face and the ear cranium angle, so as not to block the inner edge of the helix.

[0081] A recognition analysis method for collecting ear image by an ear image collection device, comprising the following steps:

[0082] Step 1: The ear image photos obtained by standardized shooting are segmented into auricle contour and face part, and the auricle inclination angle is corrected according to the auricle mark line, to obtain the vector of the initial ear shape of the ear image (reference Figure 6 ).

[0083] Step 2: Establishing ear point partition, auricle physiological characteristics and contour point, ear point pathological characteristic sample library, realizing automatic recognition and positioning of auricle degree area on the basis of labeled sample ear image and machine learning training.

[0084] Firstly, ear point partition, auricle physiological characteristics and auricle point recognition training are established, which are as follows:

[0085] S1: A total of 31 physiological feature points of the auricle were identified;

[0086] S2: Based on the national standard for auricular acupoints, marginal segmentation was performed to obtain 91 corresponding auricular landmark points. These landmark points were then connected to divide the auricular acupoints into different zones. According to the latest national standard for auricular acupoints GB / T13734-2008, 31 points were selected, including points with special ear shapes, the highest point of the auricle, and the lowest point of the auricle (nodes of each zone). To better represent the morphological structure of the auricle, 60 additional equally spaced points were added between these 31 feature points, resulting in a total of 91 points, which were marked sequentially (refer to Table 1 and...). Figure 7 , Figure 8 ).

[0087]

[0088] Furthermore, consistency of annotation is a crucial aspect of the annotation process; the consistency of annotations by different annotators measures the uniformity of the key points they annotate. Since there are as many as 91 key points to annotate, a large number of people are required to annotate, and the consistency must be evaluated. Preferably, four people (A, B, C, and D) are used for annotation. A professional TCM doctor first provides systematic training to these four individuals, and during the annotation process, the TCM doctor verifies and evaluates the annotations.

[0089] The consistency evaluation of annotations is as follows:

[0090] Within-group correlation coefficient (ICC) was used for credit rating. ICC is one of the indicators used to measure and evaluate inter-observer reliability. ICC equals the individual variability divided by the total variability, so its value ranges from 0 to 1, where 0 indicates unreliable and 1 indicates completely reliable. Generally, an ICC coefficient below 0.4 is considered to indicate poor reliability, while a coefficient above 0.75 is considered to indicate high reliability. The ICC calculation formula is as follows: ICC = In the above formula The mean square (i.e., variance) of the observed object. The mean square of the error. Let k be the mean square of the number of observers, k be the number of repetitions, and n be the number of observed subjects.

[0091] The annotation and evaluation process involved four annotators (A, B, C, and D) jointly annotating eight images, including four images each of the left and right ears. Due to the large number of key annotation points, the ear width-to-length ratio was chosen as the primary factor. Figure 8 The line connecting points 14 and 22 is used as the consistency evaluation content. The calculation results are shown in Table 2:

[0092] From the above table, the final ICC correlation coefficient value is 0.963 (95% CI: 0.897~0.992), and the ICC intra-class correlation coefficient value is higher than 0.9, which means that the labeling has high consistency.

[0093] S3: Secondly, the labeled data is trained and recognized, specifically as follows:

[0094] The Shape model is established, and M pictures are used as training samples, and each picture is labeled with N (91) feature point coordinates, which is expressed as: (x, y). It is stipulated that the expression form of each sample shape vector is: Each sample picture has a set of shape vectors S, and the length is 2*N. i indicates the i-th sample.

[0095] Since the size of the photographed pictures may not be consistent, the overall size of the picture is scaled according to the diagonal length of 200, keeping the length unchanged. Finally, the diagonal length of all pictures is consistent, and the feature vectors are also transformed in the same way. For example, if the width is reduced to c times the original, then the x in the shape vector is c times the original.

[0096] After the picture is preprocessed, the active shape model is trained, and the generation process is to first perform GPA, and then perform PCA, and the active shape model function is expressed as: ;

[0097] Among them, Savg is the average shape of the entire model, pi is a weight indicating how much change, and Si is the shape change, indicating the overall direction of movement of all feature points.

[0098] Further, GPA is performed, which aligns the data. The size of the ear in different pictures is different, the rotation angle is different, and the position of the ear is different, which leads to rotation, translation and scaling of the feature points in each sample. Linear transformation, such transformations should not be included in the shape model, that is, the comprehensive model of the sample should be trained by removing similar transformations.

[0099] Two samples are set, the first is the template sample, and the second is the sample to be aligned, and the following steps are used to complete the alignment of all samples to the template sample.

[0100] Delete the data translation component, and calculate the mean value of the position of all feature points for each sample , and then subtract this mean value from the overall feature point set to remove the translation component of the data.

[0101] Delete the scaling component of the data, calculate the standard deviation of the feature points in each sample, and divide the overall data by the standard deviation to obtain the data without scaling.

[0102] Delete the rotation component, find a is two sample bias minimum rotation angle, first out of the data covariance matrix, and then SVD decomposition, R = UV^T; Where, R is the rotation matrix, multiplied by the rotation matrix can be rotated to the angle of the template data.

[0103] Reference to the following mathematical proof (from Wikipedia) on the GPA in the proof of the optimization of the rotation matrix solution:

[0104]

[0105] Further, PCA principal component analysis, since the problem exists 91 feature points, each feature point has two dimensions, so the shape model exists 182 dimensions (i.e. the optimization of the entire shape model has 184 directions can move), the data dimension is too high, and the feature points and feature points between the change is not independent, in this case, PCA dimensionality reduction can extract the main difference between samples, and in the form of high-dimensional vector (182-dimensional) expression.

[0106] First, all the features have been aligned to the center, the function expression is: Where, M is the total amount of quantity, The i-th feature of the j-th sample is represented by

[0107] Generate the covariance matrix, which shows the relationship between different features. If the value corresponding to the position is large, it proves that the two features have a high degree of coincidence and strong similarity.

[0108]

[0109] Then, perform eigenvalue decomposition, take out the largest Ls eigenvectors corresponding to the largest Ls eigenvalues (S1, S2,..., Sls′), each eigenvector is 182-dimensional, representing the overall change direction of the model, and the shape model is composed of:

[0110] Where, Savg is the model obtained by averaging each aligned feature, which is also the average model, Si represents each change, and pi represents how much each change affects. (Reference Figure 9 , image comparison before and after feature training)

[0111] Step three: adopt convolution algorithm, local gray model, minimum variance comparison, optimize the graphics algorithm, complete ear feature extraction, including color features, texture features, and shape features, and automatically segment and save the image of the feature part (reference Figure 10 );

[0112] ​Step four: according to the analysis result of extracting ear image features, the features of the ear acupoint area are compared with the feature database to form a pre-diagnosis analysis report.

[0113] Through the ear image recognition analysis method, the effective recognition, segmentation and feature point matching of the ear acupoint area image can be improved, 91 feature points of the ear image are obtained, so as to recognize and train the ear acupoint area, the recognition accuracy of the ear image acupoint positioning is higher than that of the traditional method, the feature recognition more embodies the relevance with diseases and symptoms, so that more accurate analysis is obtained.

[0114] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An ear image acquisition device, characterized by: The handle and the cover body are connected with the end of the handle; The cover body is trumpet-shaped and is provided with a reflecting part, a folding part and a light-proof part from inside to outside in sequence, the front end of the folding part is provided with a standard color card ring along the circumference at the junction with the light-proof part; The light-proof part is provided with an annular scale card adjacent to the standard color card ring, the standard color card ring and the scale card are connected with the inner wall of the light-proof part by adhesive; The edge of the light-proof part is provided with two convex edges and two concave edges, the two convex edges correspond to each other, the two concave edges correspond to each other, the convex edges and the concave edges are staggered and connected with smooth curves; the radial section of the folding part is wavy, can be folded and stretched, and the folding angle is 3-5°; The handle is provided with a control switch, the inside of the handle is provided with a control circuit board, the end of the handle is provided with a lens module, the lens module is electrically connected with the control circuit board, the periphery of the lens module is provided with an annular flexible light source, the lamp bead surface of the annular flexible light source is provided with a soft light cover.

2. The ear image acquisition device of claim 1, wherein: The cover body is integrally made of soft rubber material, the inner side of the reflecting part, the folding part and the light-proof part is coated with a white reflecting coating.

3. The ear image acquisition device of claim 1, wherein: The control circuit board is provided with a wireless connection component and a wired connection component, the computer can be connected with the control circuit board in a wireless or wired manner through the wireless connection component and the wired connection component; The back of the handle is further provided with a display, the display is electrically connected with the control circuit board and can display the collected ear image information in real time.

4. The ear image acquisition device of claim 1, wherein: The cover body and the handle are detachably connected through threads or buckles.

Images