A left and right ear separate type auricle image acquisition and earprint identification analysis method

Abstract

The present application relates to a left and right ear separate collection auricle image collection and earprint identification analysis method, belongs to the field of biometric identification and digital image processing technology. Among them, the method comprises: S1 renders left and right ear separate collection guide interface and outputs corresponding operation prompt, carries out multi-dimensional quality check; S2 obtains left and right ear image data in turn, generates standardized bilateral auricle image data set; S3 extracts macro and micro features through bilateral earprint check model layer by layer, corrects recognition deviation through correlation comparison and cross validation, and outputs bilateral earprint determination result; S4 calculates quantitative confidence and starts grading processing mechanism, generates corresponding instruction, and finally generates structured identification analysis report. The efficient and accurate collection and identification of auricle image are realized, the identification accuracy and system running efficiency are optimized, the low collection efficiency problem caused by unilateral unqualified full re-collection in the prior art is effectively solved, and the method can be widely applied in medical identity verification, security access control and other fields.

Description

Technical Field

[0001] This invention belongs to the field of biometric recognition and digital image processing technology, specifically relating to a method for acquiring and analyzing auricle images and ear prints by separately acquiring left and right ear images. Background Technology

[0002] Biometric identification technology verifies identity by extracting inherent physiological characteristics of the human body. Fingerprint and facial recognition technologies have been widely used in identity authentication, security access control, and medical health. Earprints, as a unique biometric feature, have the characteristics of non-contact collection, resistance to tampering, significant individual differences, and are not affected by changes in facial expressions or makeup. They have significant advantages in scenarios with high public health and safety requirements and are gradually becoming an important research direction in the field of biometric identification.

[0003] Existing earprint acquisition technology has obvious defects. The acquisition process adopts a bilateral one-time completion mode. If the image on one side does not meet the acquisition standard, the bilateral image acquisition operation needs to be repeated. The qualified image data that has been acquired cannot be retained. Users need to repeatedly adjust the shooting posture. The average time for a single acquisition increases by more than double, which greatly reduces the user experience and is likely to cause queuing congestion in public places with high traffic.

[0004] Existing earprint recognition models mostly adopt a single-branch architecture that trains and recognizes images independently on one side. They do not establish the correlation between the features of the two sides of the earprint and cannot use the natural symmetry features of the two sides of the human earprint to correct errors. Local defects in the one-sided image will be directly transmitted to the final recognition result. When there are problems such as shooting angle deviation, uneven lighting or slight local occlusion, the recognition accuracy will drop by more than 20%.

[0005] Meanwhile, the existing image quality verification technology has a relatively simple dimension, and can only output a binary judgment result of qualified or unqualified. It cannot comprehensively evaluate multiple key dimensions of image acquisition quality. The recognition result processing flow is not refined enough, and only two processing methods are set: pass and rejection. It cannot perform corresponding differentiated processing operations according to different recognition confidence levels, which makes it difficult to balance the overall recognition accuracy and system operating efficiency, and cannot meet the needs of large-scale practical application scenarios. Summary of the Invention

[0006] To address the aforementioned problems in the existing technology, this invention provides a method for acquiring and analyzing left and right ear images of auricles and earprint recognition. The objective of this invention can be achieved through the following technical solution: include: S1: Render the left and right ear separate acquisition guidance interface, and output the corresponding shooting requirements and operation prompts respectively; perform multi-dimensional quality verification on the current ear auricle image in real-time preview, generate quality verification results, and enable the corresponding acquisition permission after the image quality meets the standard. S2: Based on the quality verification results, acquire left and right ear image data sequentially; after each side's data acquisition is completed, automatically jump to the acquisition process of the other side and output a bilateral auricle image dataset that has passed quality verification; S3: Perform image preprocessing on the bilateral auricle image dataset, segment the auricle contours of each side and locate the feature regions; through the pre-constructed bilateral auricle verification model, extract macroscopic texture features and microscopic morphological features in layers, and perform correlation comparison and cross-validation to correct the recognition bias introduced by the difference in shooting angle or lighting on one side, and output the bilateral auricle judgment result. S4: Based on the bilateral earprint determination results, calculate the quantitative confidence level of the bilateral texture recognition results; activate the hierarchical result output and anomaly handling mechanism, and generate corresponding processing instructions based on the comparison results of the confidence level of each side with the preset threshold; aggregate the texture type and confidence level information of the bilateral auricle images to generate a structured earprint recognition analysis report.

[0007] Specifically, the process of outputting the corresponding shooting requirements and operation prompts is as follows: Automatically identify the side of the auricle to be captured, render the corresponding semi-transparent standard frame reference line overlaid on the real-time preview interface, and dynamically output shooting distance adjustment prompts to guide the user to fully align the auricle with the reference line. Simultaneously, a standard shooting posture diagram is rendered to show the correct relative position of the head and the device, matching the preset user group classification, and retrieving the corresponding prompt text font size and display duration parameters according to the preset group and parameter mapping relationship, and updating the prompt content in real time.

[0008] Specifically, the process of performing multi-dimensional quality verification on the real-time preview of the current lateral auricle image is as follows: Image data is extracted frame by frame from the real-time preview stream, and the edge detection algorithm is used to extract the ear edge contour to determine whether the ear is completely in the frame. The uniformity of light is assessed by statistically analyzing the grayscale histogram distribution of the image, the relative offset between the center point of the ear canal and the geometric center of the auricle is calculated to determine the shooting angle, and the detection results are compared with preset standards dimensionally to generate the corresponding quality judgment results in real time. The quality feedback information is simultaneously rendered to the specified area of ​​the preview interface.

[0009] Specifically, the process of outputting the quality-verified bilateral auricle image dataset is as follows: The system performs side-by-side acquisition process control on the acquired left and right ear image data. After the data acquisition of each side is completed, it automatically jumps to the acquisition process of the other side. For data that does not meet the preset requirements on one side, a re-acquisition operation is triggered, and the image data of the other side that has met the preset requirements is retained. The final acquired image data is standardized to unify the image resolution and color space, remove invalid and redundant information from the image, and associate each image with the corresponding acquisition side, quality verification result and acquisition timestamp. The data is organized according to a preset hierarchical structure to generate a complete dataset containing bilateral images and associated metadata.

[0010] Specifically, the bilateral earprint verification model includes an input layer, a feature extraction layer, a cross-validation layer, and an output layer, and its construction process is as follows: The input layer receives standardized auricle image data after preprocessing from both sides and performs format conversion. The feature extraction layer performs multi-scale convolution operations on the left and right ear images to extract basic features at different levels. The cross-validation layer establishes the corresponding relationship between the features of both sides and uses symmetry to correct errors. The output layer performs classification operations on the validated features based on the extracted feature vectors and outputs the texture type determination result and matching degree value of the two ear prints.

[0011] Specifically, the process of segmenting the contours of each auricle and locating the feature regions is as follows: Semantic segmentation is performed on the preprocessed auricle image to extract the binary mask of the auricle region, remove interference from irrelevant regions, and extract the continuous and smooth outer contour line of the auricle through an edge tracking algorithm. Based on the anatomical structure of the auricle and the distribution standards of auricular acupoints in traditional Chinese medicine, key functional areas including the concha, helix, earlobe and antihelix are divided, and feature areas containing core auricular texture are located.

[0012] Specifically, the process of extracting macroscopic texture features and microscopic morphological features in a layered manner is as follows: By extracting feature maps of different resolutions through a multi-scale convolutional network, the image feature maps of different resolutions are output in layers, and macroscopic texture features and microscopic morphological features are extracted from the feature maps of each layer respectively. Normalization preprocessing is performed on all extracted feature data to eliminate the differences in data dimensions of different feature dimensions, and to uniformly map heterogeneous feature data into standardized feature vectors with fixed dimensions and fixed lengths.

[0013] Specifically, the process of outputting the bilateral earprint determination results is as follows: The preprocessed bilateral auricle images are input into the bilateral auricle verification model, and feature extraction, cross-validation and classification operations are performed in sequence. The matching degree between the features of each side and all types in the preset texture feature library is calculated respectively. The matching scores are sorted from high to low, and the texture type with the highest matching score is selected as the preliminary judgment result for the corresponding side. The final matching score is adjusted by combining the error correction value output by the cross-validation layer, and the judgment results of the left and right ears are integrated to form a complete bilateral earprint judgment result.

[0014] Specifically, the process for calculating the quantization confidence of the bilateral texture recognition results is as follows: The comprehensive quality score, feature matching score, and texture saliency score of the two images are obtained respectively, and the corresponding weight coefficients are set according to the degree of influence of different dimensions on the recognition results. The three scores are weighted and summed to obtain the quantitative confidence level of the corresponding lateral recognition result. The confidence level calculation process for the left and right ears is executed sequentially to generate an evaluation result containing bilateral confidence level values ​​and corresponding levels.

[0015] Specifically, the process of initiating the hierarchical result output and exception handling mechanism is as follows: Read the quantized confidence scores of the two-sided recognition results and compare the confidence scores with multiple preset thresholds one by one; If the confidence level meets the first preset threshold, proceed to the report generation stage; if the confidence level meets the second preset threshold, mark the specific content items that need to be manually reviewed; if the confidence level meets the third preset threshold, start the exception handling process. Guide users to perform targeted re-collection operations, and simultaneously push the processing status information to the front-end interface for display.

[0016] Specifically, the process of generating the corresponding processing instructions is as follows: Based on the output of the hierarchical results and the judgment results of the exception handling mechanism, the corresponding execution instructions are generated; For the first preset threshold result, a report generation instruction carrying complete identification data is generated; for the second preset threshold result, a manual review instruction marking items to be reviewed is generated; and for the third preset threshold result, a re-collection instruction confirming unqualified dimensions and adjustment directions is generated.

[0017] Specifically, the process of generating the structured earprint recognition analysis report is as follows: Extract the original images of both auricles, the ear print texture type determination results, the quantitative confidence assessment results, and the abnormal prompt information generated during the acquisition process; Load the preset report template file, parse the data placeholders in the template, map and fill the extracted information into the corresponding placeholder positions in the template one by one, arrange the data blocks according to the preset content order, and generate a structured report file that conforms to the standard format. Write the report file to the specified storage path and push the report data to the front-end interface for rendering and display.

[0018] The beneficial effects of this invention are as follows: (1) By setting up a mechanism for independent acquisition of left and right ears and single-sided re-acquisition, the process of automatically jumping to the acquisition process of the other side is realized after the acquisition of the image of each ear is completed. For the single-sided image that does not meet the preset requirements, only the corresponding side is re-captured, and the image data of the other side that meets the requirements is completely retained. This effectively avoids the problem of having to re-acquire all images if one side is not qualified in the existing technology, greatly shortens the average acquisition time per acquisition, improves the user experience, and is suitable for public application scenarios with large traffic flow. (2) By setting up an earprint recognition model with cross-validation of bilateral features, the corresponding relationship between bilateral features is established based on the extraction of multi-scale features of the left and right ears respectively. The error is corrected by utilizing the natural symmetry features of the bilateral earprints of the human body. This effectively offsets the recognition error caused by the deviation of the shooting angle on one side, uneven lighting or slight local occlusion, and significantly improves the overall accuracy and robustness of earprint recognition. At the same time, combined with the multi-dimensional image quality verification and graded result processing flow, the recognition accuracy and system operating efficiency are improved simultaneously. Attached Figure Description

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0020] Figure 1 This is a flowchart illustrating the left and right ear image acquisition and earprint recognition analysis method of the present invention. Figure 2 This is a data flow diagram of the left and right ear image acquisition and ear print recognition analysis method of the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0022] Please see Figures 1-2 A method for acquiring and analyzing auricle images and earprint recognition using a method that separates left and right ear images; include: S1: Render the left and right ear separate acquisition guidance interface, and output the corresponding shooting requirements and operation prompts respectively; perform multi-dimensional quality verification on the current ear auricle image in real-time preview, generate quality verification results, and enable the corresponding acquisition permission after the image quality meets the standard. S2: Based on the quality verification results, acquire left and right ear image data sequentially; after each side's data acquisition is completed, automatically jump to the acquisition process of the other side and output a bilateral auricle image dataset that has passed quality verification; S3: Perform image preprocessing on the bilateral auricle image dataset, segment the auricle contours of each side and locate the feature regions; through the pre-constructed bilateral auricle verification model, extract macroscopic texture features and microscopic morphological features in layers, and perform correlation comparison and cross-validation to correct the recognition bias introduced by the difference in shooting angle or lighting on one side, and output the bilateral auricle judgment result. S4: Based on the bilateral earprint determination results, calculate the quantitative confidence level of the bilateral texture recognition results; activate the hierarchical result output and anomaly handling mechanism, and generate corresponding processing instructions based on the comparison results of the confidence level of each side with the preset threshold; aggregate the texture type and confidence level information of the bilateral auricle images to generate a structured earprint recognition analysis report.

[0023] In this embodiment, the quality verification result refers to the judgment conclusion generated after performing multi-dimensional detection on the real-time previewed auricle image, specifically including: the judgment result of whether the auricle is completely in the frame, the evaluation result of the overall uniformity of the image lighting, the judgment result of whether the shooting angle meets the requirements, and the specific quantitative score of each dimension detection. Left and right ear image data refers to the original visual data of one ear auricle captured through acquisition permissions, specifically including: complete pixel matrix data of the original left ear image, complete pixel matrix data of the original right ear image, and basic metadata at the time of image acquisition (acquisition timestamp, acquisition device number). Macro-texture features refer to feature data that reflect the overall morphological structure of the auricle, extracted from the corresponding layers of the multi-scale image pyramid. Specifically, these include: the overall extension direction data of the main folds on the surface of the auricle, the number and distribution density data of the main folds per unit area, the depth and curvature change data of the folds, and the geometric shape data of the overall contour of the auricle. Microscopic morphological features refer to feature data that reflect the local fine structure of the auricle extracted from the corresponding layers of the multi-scale image pyramid. Specifically, these include: the location, size, and depth distribution data of pigmented spots in the local area; the spacing and direction variation data of fine textures on the skin surface; the distribution morphology data of pores and sebaceous glands; and the fine structural differences data of micro-scars and skin folds. The texture type of bilateral auricle images refers to the standardized classification conclusions output by the bilateral auricle verification model, specifically including: the standardized classification label corresponding to the left ear print, the standardized classification label corresponding to the right ear print, the unique numerical code corresponding to each classification label, and the preliminary matching degree value of the classification result. Confidence information refers to the quantitative and judgment information related to the reliability of the recognition results, specifically including: the 0-1 interval quantitative confidence value of the left ear recognition result, the 0-1 interval quantitative confidence value of the right ear recognition result, the comparison results of the confidence of each side with three preset thresholds, and the processing flow identifier corresponding to the confidence. Image data refers to all visual-related data involved in the entire process of this invention, specifically including: continuous frame-by-frame image data at 30 frames per second in the real-time preview stream, single-frame high-definition raw image data captured after the acquisition permission is enabled, image data after format standardization processing, and intermediate feature map data generated during feature extraction. Standardized auricle image data refers to auricle image data processed in a unified format, specifically including: image pixel data uniformly adjusted to a fixed resolution (such as 1080×1080 pixels), image data uniformly converted to the sRGB standard color space, image data cropped to remove irrelevant background areas outside the auricle, and image data to remove compression artifacts and noise; Core ear print texture refers to the key ear print texture area with the highest individual recognizability and used for identity recognition. Specifically, it includes: the fold texture structure of the concha cavity and cymba conchae region, the curl texture structure of the helix edge and crus of the helix, the bifurcation texture structure of the antihelix and antihelix crus, and the surface texture and morphological structure of the earlobe region. The quantitative confidence assessment result refers to the numerical conclusion reflecting the reliability of the recognition result obtained through multi-dimensional weighted calculation. Specifically, it includes: the comprehensive quality score of the unilateral image (obtained by weighting the scores of completeness, lighting, and angle), the matching score between the unilateral feature and the feature library, the significance score of the unilateral ear print texture, and the final confidence value after weighted summation of the three scores. Abnormal prompts refer to prompts generated during the data acquisition and recognition process to guide user operation or manual review. Specifically, they include: specific dimensions of image quality failure (such as "ear not fully in the frame" or "light too dim"), side and feature area markings for recognition results requiring manual review, specific adjustment directions for re-acquisition (such as "please turn your head 15 degrees to the right"), and alarm information for abnormal system status.

[0024] Specifically, the process of outputting the corresponding shooting requirements and operation prompts is as follows: Automatically identify the side of the auricle to be captured, render the corresponding semi-transparent standard frame reference line overlaid on the real-time preview interface, and dynamically output shooting distance adjustment prompts to guide the user to fully align the auricle with the reference line. Simultaneously, a standard shooting posture diagram is rendered to show the correct relative position of the head and the device, matching the preset user group classification, and retrieving the corresponding prompt text font size and display duration parameters according to the preset group and parameter mapping relationship, and updating the prompt content in real time.

[0025] Specifically, the process of performing multi-dimensional quality verification on the real-time preview of the current lateral auricle image is as follows: Image data is extracted frame by frame from the real-time preview stream, and the edge detection algorithm is used to extract the ear edge contour to determine whether the ear is completely in the frame. The uniformity of light is assessed by statistically analyzing the grayscale histogram distribution of the image, the relative offset between the center point of the ear canal and the geometric center of the auricle is calculated to determine the shooting angle, and the detection results are compared with preset standards dimensionally to generate the corresponding quality judgment results in real time. The quality feedback information is simultaneously rendered to the specified area of ​​the preview interface.

[0026] Specifically, the process of outputting the quality-verified bilateral auricle image dataset is as follows: The system performs side-by-side acquisition process control on the acquired left and right ear image data. After the data acquisition of each side is completed, it automatically jumps to the acquisition process of the other side. For data that does not meet the preset requirements on one side, a re-acquisition operation is triggered, and the image data of the other side that has met the preset requirements is retained. The final acquired image data is standardized to unify the image resolution and color space, remove invalid and redundant information from the image, and associate each image with the corresponding acquisition side, quality verification result and acquisition timestamp. The data is organized according to a preset hierarchical structure to generate a complete dataset containing bilateral images and associated metadata.

[0027] Specifically, the bilateral earprint verification model includes an input layer, a feature extraction layer, a cross-validation layer, and an output layer, and its construction process is as follows: The input layer receives standardized auricle image data after preprocessing from both sides and performs format conversion. The feature extraction layer performs multi-scale convolution operations on the left and right ear images to extract basic features at different levels. The cross-validation layer establishes the corresponding relationship between the features of both sides and uses symmetry to correct errors. The output layer performs classification operations on the validated features based on the extracted feature vectors and outputs the texture type determination result and matching degree value of the two ear prints.

[0028] Specifically, the process of segmenting the contours of each auricle and locating the feature regions is as follows: Semantic segmentation is performed on the preprocessed auricle image to extract the binary mask of the auricle region, remove interference from irrelevant regions, and extract the continuous and smooth outer contour line of the auricle through an edge tracking algorithm. Based on the anatomical structure of the auricle and the distribution standards of auricular acupoints in traditional Chinese medicine, key functional areas including the concha, helix, earlobe and antihelix are divided, and feature areas containing core auricular texture are located.

[0029] Specifically, the process of extracting macroscopic texture features and microscopic morphological features in a layered manner is as follows: By extracting feature maps of different resolutions through a multi-scale convolutional network, the image feature maps of different resolutions are output in layers, and macroscopic texture features and microscopic morphological features are extracted from the feature maps of each layer respectively. Normalization preprocessing is performed on all extracted feature data to eliminate the differences in data dimensions of different feature dimensions, and to uniformly map heterogeneous feature data into standardized feature vectors with fixed dimensions and fixed lengths.

[0030] Specifically, the process of outputting the bilateral earprint determination results is as follows: The preprocessed bilateral auricle images are input into the bilateral auricle verification model, and feature extraction, cross-validation and classification operations are performed in sequence. The matching degree between the features of each side and all types in the preset texture feature library is calculated respectively. The matching scores are sorted from high to low, and the texture type with the highest matching score is selected as the preliminary judgment result for the corresponding side. The final matching score is adjusted by combining the error correction value output by the cross-validation layer, and the judgment results of the left and right ears are integrated to form a complete bilateral earprint judgment result.

[0031] Specifically, the process for calculating the quantization confidence of the bilateral texture recognition results is as follows: The comprehensive quality score, feature matching score, and texture saliency score of the two images are obtained respectively, and the corresponding weight coefficients are set according to the degree of influence of different dimensions on the recognition results. The three scores are weighted and summed to obtain the quantitative confidence level of the corresponding lateral recognition result. The confidence level calculation process for the left and right ears is executed sequentially to generate an evaluation result containing bilateral confidence level values ​​and corresponding levels.

[0032] Specifically, the process of initiating the hierarchical result output and exception handling mechanism is as follows: Read the quantized confidence scores of the two-sided recognition results and compare the confidence scores with multiple preset thresholds one by one; If the confidence level meets the first preset threshold, proceed to the report generation stage; if the confidence level meets the second preset threshold, mark the specific content items that need to be manually reviewed; if the confidence level meets the third preset threshold, start the exception handling process. Guide users to perform targeted re-collection operations, and simultaneously push the processing status information to the front-end interface for display.

[0033] Specifically, the process of generating the corresponding processing instructions is as follows: Based on the output of the hierarchical results and the judgment results of the exception handling mechanism, the corresponding execution instructions are generated; For the first preset threshold result, a report generation instruction carrying complete identification data is generated; for the second preset threshold result, a manual review instruction marking items to be reviewed is generated; and for the third preset threshold result, a re-collection instruction confirming unqualified dimensions and adjustment directions is generated.

[0034] Specifically, the process of generating the structured earprint recognition analysis report is as follows: Extract the original images of both auricles, the ear print texture type determination results, the quantitative confidence assessment results, and the abnormal prompt information generated during the acquisition process; Load the preset report template file, parse the data placeholders in the template, map and fill the extracted information into the corresponding placeholder positions in the template one by one, arrange the data blocks according to the preset content order, and generate a structured report file that conforms to the standard format. Write the report file to the specified storage path and push the report data to the front-end interface for rendering and display.

[0035] In this embodiment, the specific process of rendering the left and right ear side acquisition guidance interface is as follows: automatically identify the current side to be acquired, render the corresponding side's semi-transparent white standard reference frame, set the reference line width to 2 pixels and the transparency to 0.7, and overlay it in the center of the real-time preview interface; dynamically output shooting distance adjustment prompts, displaying a green prompt when the shooting distance is within the range of 30-50cm, and a red prompt when it exceeds the range; at the same time, render a standard shooting posture diagram in the lower right corner of the interface.

[0036] In this embodiment, the specific implementation process of adaptively adjusting the size and display duration of the prompt text for different user groups is as follows: The terminal reads the patient age tag transmitted from the hospital HIS system or the user classification identifier preset locally on the terminal, and matches it with the preset three user group categories: ordinary adults, the elderly, and children; it loads the locally stored group-parameter mapping configuration table, in which the prompt text font size for ordinary adults is 16 and the display duration is 2 seconds, the prompt text font size for the elderly is 24 and the display duration is 5 seconds, and the prompt text font size for the children is 20 and the display duration is 3 seconds; the corresponding parameters are automatically retrieved and the prompt content is rendered according to the matched user group. When the user is switched or the user identity tag is updated during the data collection process, the display parameters of the prompt text are refreshed in real time; if no valid user identity tag is identified, the parameter configuration corresponding to ordinary adults is used by default.

[0037] In this embodiment, the specific process of multi-dimensional quality verification of real-time preview images is as follows: Image data at 30fps is extracted frame by frame from the real-time preview stream; the Canny edge detection algorithm is used to extract the ear's edge contour; high and low thresholds are set to 80 and 200 respectively; edge closure is calculated, and a closure ≥ 0.95 indicates the ear is completely within the frame; the grayscale histogram variance method is used to evaluate the uniformity of light, and a variance ≤ 50 indicates the light is acceptable; the relative offset between the center point of the ear canal and the geometric center of the ear is calculated, and a horizontal offset ≤ 15 pixels and a vertical offset ≤ 10 pixels indicate the shooting angle is acceptable; the verification time for each frame is controlled within 15ms, and quality feedback information is rendered in real-time to the upper left corner of the preview interface.

[0038] In this embodiment, the specific process of side-by-side acquisition process control and dataset generation is as follows: independent acquisition process control is performed on the left and right ear images. After the data capture of each side is completed, the process automatically jumps to the acquisition process of the other side. If the image quality of a single side is unqualified, only the recapture operation of that side is triggered, and the qualified image data of the other side is completely retained. The finally acquired images are subjected to format standardization processing, with a unified resolution of 1080×1080 pixels and a unified color space of sRGB. The background area outside the auricle is cropped and removed. Metadata such as acquisition side, quality verification score, acquisition timestamp and device number are associated with each side image. The data is organized according to the hierarchical structure of "acquisition date / device number / side" to generate a complete bilateral auricle image dataset.

[0039] In this embodiment, the specific process of auricular contour segmentation and core feature region localization is as follows: The U-Net semantic segmentation network is used to extract the binary mask of the auricular region. The network input size is 512×512 pixels, and the output is a binary segmentation map of the same size. The intersection-union ratio threshold is set to 0.9. The Suzuki85 edge tracking algorithm is used to extract the continuous and smooth outer contour line of the auricle and remove discrete edge noise with a length of less than 50 pixels. Based on the anatomical structure of the auricle and the distribution standard of auricular acupoints in traditional Chinese medicine, the auricle is divided into four core functional regions: concha, helix, auricle, and antihelix. Masks for each region are generated and merged to obtain a feature region mask containing the core auricular texture.

[0040] In this embodiment, the specific process of constructing and inferring the bilateral earprint verification model is as follows: a four-layer cascaded bilateral earprint verification model is constructed. The input layer receives standardized image data of 1080×1080×3 and converts it into a 32-bit floating-point tensor. The feature extraction layer adopts an improved ResNet-18 network with three parallel convolutional branches. Convolutional kernels of 3×3, 5×5, and 7×7 are used to perform multi-scale feature extraction, and the output is a 512-dimensional basic feature vector. The cross-validation layer calculates the cosine similarity between the feature vectors of the left and right ears, establishes a one-to-one correspondence between 128 feature points, and calculates the error correction value using bilateral symmetry. The correction coefficient is set to 0.7. The output layer adopts a Softmax classifier and outputs the probability distribution of 128 preset earprint texture types and the matching degree value in the 0-1 interval. The model training batch size is set to 32, the number of iterations is set to 100 rounds, and the learning rate is set to 0.001.

[0041] In this embodiment, the specific process of multi-scale hierarchical feature extraction and normalization is as follows: A fourth-order Gaussian image pyramid decomposition is performed on the located core feature region, and feature maps at four resolution levels (540×540, 270×270, 135×135, and 67×67) are generated sequentially through downsampling. Macroscopic texture features such as ear fold direction, density, and unevenness are extracted from the 540×540 and 270×270 levels, while microscopic morphological features such as pigmentation, fine texture, and skin morphology are extracted from the 135×135 and 67×67 levels. All features are normalized using the Z-score normalization method, mapping features of different dimensions to values ​​with a mean of 0 and a variance of 1, and finally concatenating them to generate a 256-dimensional normalized feature vector.

[0042] In this embodiment, the specific process of quantization confidence calculation and graded result processing is as follows: Calculate the comprehensive quality score, feature matching score, and texture saliency score of the single-sided image respectively, setting weight coefficients of 0.3, 0.5, and 0.2 respectively, and sum them by weight to obtain the quantization confidence score in the 0-1 interval; preset three confidence thresholds: the first preset threshold is 0.85, the second preset threshold is 0.6, and the third preset threshold is 0.3; if the confidence score is ≥0.85, directly jump to the report generation stage; if the confidence score is 0.6≤confidence score<0.85, mark the side and feature regions that need manual review; if the confidence score is <0.3, start the anomaly handling process and push a re-collection prompt containing specific adjustment directions to the front-end interface.

[0043] In this embodiment, the specific process of generating and outputting the structured earprint recognition analysis report is as follows: extracting the original images of both auricles, texture type determination results, quantified confidence scores, and abnormal prompt information; loading a preset XML format report template, parsing the data placeholders in the template, and mapping and filling the extracted information into the corresponding positions one by one; arranging the data blocks in the order of the collected information area, the bilateral image display area, the texture feature analysis area, the confidence assessment area, and the abnormal prompt area; and simultaneously generating a PDF format human-reading report and a JSON format machine-parsed report.

[0044] In this embodiment, taking the patient identity verification and consultation process in the ENT department of a hospital outpatient clinic as an example, the specific implementation process is as follows: A represents the optimal distance range between the acquisition device and the patient's ear; B represents the maximum allowable time for single-frame image quality verification; G represents the preset standard image resolution; H represents the industry-standard color space; C represents the average inference time for a single earprint recognition; I represents the preset fixed dimension of the feature vector; J represents the weighting coefficient for bilateral feature error correction; D represents the direct pass threshold; E represents the manual review threshold; F represents the re-acquisition threshold; and K represents the specified storage path for patient identity and medical record data. S1: The verification terminal at the clinic entrance automatically wakes up and renders the left and right ear side acquisition guidance interface, outputting corresponding shooting requirements and operation prompts according to the current side to be acquired; it performs multi-dimensional quality verification on the real-time preview of the current ear image, generates quality verification results, and enables the corresponding acquisition permission after the image quality meets the standards; specifically: after the patient completes registration and arrives at the clinic entrance, the terminal reads the patient's medical card information and starts the acquisition program, first rendering the acquisition interface for the corresponding side, overlaying a semi-transparent standard frame reference line, and dynamically prompting the patient to adjust the head position and shooting distance to range A; it extracts image data from the real-time preview stream frame by frame, uses the Canny edge detection algorithm to determine the complete frame of the ear, the grayscale histogram variance method to evaluate the uniformity of light, and the geometric center offset method to determine the shooting angle, with the single-frame verification time controlled within B, and automatically enables the acquisition permission after all dimensions of verification meet the standards; S2: Based on the quality verification results, acquire left and right ear image data sequentially; after each side's data acquisition is completed, automatically jump to the acquisition process of the other side, and output a quality-verified bilateral auricle image dataset; specifically: after the acquisition permission is enabled, automatically capture the high-definition image of the current side, and then automatically jump to the acquisition process of the other side; if the image of one side does not meet the quality standard, only trigger the recapture operation of that side, and completely retain the qualified image data of the other side; perform format standardization processing on the finally acquired bilateral images, unify the resolution to G and the color space to H, and crop and remove irrelevant background areas outside the auricle; associate each image with the patient's unique identifier, consultation number, acquisition timestamp and terminal number metadata, organize the data according to the preset hierarchical structure, and generate a complete bilateral auricle image dataset; S3: Perform image preprocessing on the bilateral auricle image dataset, segment the auricle contours of each side and locate feature regions; call the pre-built bilateral auricle verification model, extract macroscopic texture features and microscopic morphological features layer by layer, and perform correlation comparison and cross-validation to correct the recognition bias introduced by the difference in shooting angle or lighting on one side, and output the bilateral auricle judgment result; Specifically: the terminal uses the U-Net semantic segmentation network to extract the binary mask of the auricle region, extracts the continuous auricle contour through the Suzuki85 edge tracking algorithm, and divides the concha, helix, and other 4 core feature regions based on anatomical standards; upload the standardized bilateral images to the hospital cloud server, call the pre-trained bilateral auricle verification model and perform matching inference with the patient auricle templates pre-stored in the HIS system, and the average inference time of the model is C; inside the model, I-dimensional basic feature vectors are extracted through multi-scale convolution operation, the cross-validation layer establishes the correspondence between bilateral feature points, the error correction value is calculated using symmetry, the correction coefficient is set to J, and finally the bilateral auricle matching result and matching degree value are output; S4: Calculate the quantitative confidence of the bilateral texture recognition results based on the bilateral ear print determination results; start a hierarchical result output and exception handling mechanism, and generate corresponding processing instructions according to the comparison results between the confidence of each side and the preset threshold; aggregate the bilateral auricle image texture type and confidence information to generate a structured ear print recognition analysis report; specifically, the cloud end separately calculates the comprehensive quality score, feature matching score and texture saliency score of the single-side image, and obtains the quantitative confidence in the 0-1 interval by weighted summation according to the preset weights; compare the confidence value with the three preset thresholds D, E and F. If the confidence is ≥ D, directly trigger the consultation room calling system and display the patient's consultation serial number; if E ≤ confidence < D, push the recognition result to the nurse station background for manual review; if the confidence is < F, return to the terminal to prompt the patient to adjust in the specified direction and perform re-collection; after the verification is completed, a structured report in both PDF and JSON formats is automatically generated, stored in the K path of the server and associated with the patient's electronic medical record, and simultaneously synchronized to the doctor's workstation in the consultation room, supporting subsequent consultation traceability and data statistics.

[0045] The above are only preferred embodiments of the present invention, and are not intended to limit the present invention in any form. Although the present invention has been disclosed above by the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art, without departing from the scope of the technical solution of the present invention, may make some changes or modifications using the technical content disclosed above to make equivalent embodiments with equivalent changes. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A method for acquiring and analyzing left and right ear images of auricles and earprint recognition, characterized in that, include: S1: Render the left and right ear side acquisition guidance interface, and output the corresponding shooting requirements and operation prompts respectively; Perform multi-dimensional quality checks on the current side ear image in real-time preview, generate quality check results, and enable the corresponding acquisition permissions only after the image quality meets the standards. S2: Based on the quality verification results, acquire left and right ear image data sequentially; after each side's data acquisition is completed, automatically jump to the acquisition process of the other side and output a bilateral auricle image dataset that has passed quality verification; S3: Perform image preprocessing on the bilateral auricle image dataset, segment the auricle contours of each side and locate the feature regions; through the pre-constructed bilateral auricle verification model, extract macroscopic texture features and microscopic morphological features in layers, and perform correlation comparison and cross-validation to correct the recognition bias introduced by the difference in shooting angle or lighting on one side, and output the bilateral auricle judgment result. S4: Based on the bilateral earprint determination results, calculate the quantized confidence level of the bilateral texture recognition results; The hierarchical result output and anomaly handling mechanism is activated, and corresponding processing instructions are generated based on the comparison results of the confidence levels of each side with the preset thresholds. By aggregating texture type and confidence information from bilateral auricle images, a structured earprint recognition and analysis report is generated.

2. The method according to claim 1, characterized in that, The specific process of outputting the corresponding shooting requirements and operation prompts is as follows: Automatically identify the side of the auricle to be captured, render the corresponding semi-transparent standard frame reference line overlaid on the real-time preview interface, and dynamically output shooting distance adjustment prompts to guide the user to fully align the auricle with the reference line. Simultaneously, a standard shooting posture diagram is rendered to show the correct relative position of the head and the device, matching the preset user group classification, and retrieving the corresponding prompt text font size and display duration parameters according to the preset group and parameter mapping relationship, and updating the prompt content in real time.

3. The method according to claim 1, characterized in that, The specific process of performing multi-dimensional quality verification on the real-time preview of the current side auricle image is as follows: Image data is extracted frame by frame from the real-time preview stream, and the edge detection algorithm is used to extract the ear edge contour to determine whether the ear is completely in the frame. The uniformity of light is assessed by statistically analyzing the grayscale histogram distribution of the image, the relative offset between the center point of the ear canal and the geometric center of the auricle is calculated to determine the shooting angle, and the detection results are compared with preset standards dimensionally to generate the corresponding quality judgment results in real time. The quality feedback information is simultaneously rendered to the specified area of ​​the preview interface.

4. The method according to claim 1, characterized in that, The specific process for outputting the quality-verified bilateral auricle image dataset is as follows: The system performs side-by-side acquisition process control on the acquired left and right ear image data. After the data acquisition of each side is completed, it automatically jumps to the acquisition process of the other side. For data that does not meet the preset requirements on one side, a re-acquisition operation is triggered, and the image data of the other side that has met the preset requirements is retained. The final acquired image data is standardized to unify the image resolution and color space, remove invalid and redundant information from the image, and associate each image with the corresponding acquisition side, quality verification result and acquisition timestamp. The data is organized according to a preset hierarchical structure to generate a complete dataset containing bilateral images and associated metadata.

5. The method according to claim 1, characterized in that, The bilateral earprint verification model includes an input layer, a feature extraction layer, a cross-validation layer, and an output layer. The specific construction process is as follows: The input layer receives standardized auricle image data after preprocessing from both sides and performs format conversion. The feature extraction layer performs multi-scale convolution operations on the left and right ear images to extract basic features at different levels. The cross-validation layer establishes the corresponding relationship between the features of both sides and uses symmetry to correct errors. The output layer performs classification operations on the validated features based on the extracted feature vectors and outputs the texture type determination result and matching degree value of the two ear prints.

6. The method according to claim 1, characterized in that, The specific process of segmenting the contours of each auricle and locating the feature regions is as follows: Semantic segmentation is performed on the preprocessed auricle image to extract the binary mask of the auricle region, remove interference from irrelevant regions, and extract the continuous and smooth outer contour line of the auricle through an edge tracking algorithm. Based on the anatomical structure of the auricle and the distribution standards of auricular acupoints in traditional Chinese medicine, key functional areas including the concha, helix, earlobe and antihelix are divided, and feature areas containing core auricular texture are located.

7. The method according to claim 1, characterized in that, The specific extraction process of the macroscopic texture features and microscopic morphological features is as follows: By extracting feature maps of different resolutions through a multi-scale convolutional network, the image feature maps of different resolutions are output in layers, and macroscopic texture features and microscopic morphological features are extracted from the feature maps of each layer respectively. Normalization preprocessing is performed on all extracted feature data to eliminate the differences in data dimensions of different feature dimensions, and to uniformly map heterogeneous feature data into standardized feature vectors with fixed dimensions and fixed lengths.

8. The method according to claim 1, characterized in that, The specific output process of the bilateral earprint determination results is as follows: The preprocessed bilateral auricle images are input into the bilateral auricle verification model, and feature extraction, cross-validation and classification operations are performed in sequence. The matching degree between the features of each side and all types in the preset texture feature library is calculated respectively. The matching scores are sorted from high to low, and the texture type with the highest matching score is selected as the preliminary judgment result for the corresponding side. The final matching score is adjusted by combining the error correction value output by the cross-validation layer, and the judgment results of the left and right ears are integrated to form a complete bilateral earprint judgment result.

9. The method according to claim 1, characterized in that, The specific calculation process for the quantization confidence of the bilateral texture recognition results is as follows: The comprehensive quality score, feature matching score, and texture saliency score of the two images are obtained respectively, and the corresponding weight coefficients are set according to the degree of influence of different dimensions on the recognition results. The three scores are weighted and summed to obtain the quantitative confidence level of the corresponding lateral recognition result. The confidence level calculation process for the left and right ears is executed sequentially to generate an evaluation result containing bilateral confidence level values ​​and corresponding levels.

10. The method according to claim 1, characterized in that, The specific activation process of the hierarchical result output and anomaly handling mechanism is as follows: Read the quantified confidence value of the two-sided recognition results, compare the confidence value with multiple preset thresholds one by one. If the confidence value meets the first preset threshold, jump to the report generation stage. If the confidence value meets the second preset threshold, mark the specific content items that need to be manually reviewed. If the confidence value meets the third preset threshold, start the exception handling process. Guide users to perform targeted re-collection operations, and simultaneously push the processing status information to the front-end interface for display.

11. The method according to claim 1, characterized in that, The specific generation process of the corresponding processing instruction is as follows: Based on the output of the hierarchical results and the judgment results of the exception handling mechanism, the corresponding execution instructions are generated; For the first preset threshold result, a report generation instruction carrying complete identification data is generated; for the second preset threshold result, a manual review instruction marking items to be reviewed is generated; and for the third preset threshold result, a re-collection instruction confirming unqualified dimensions and adjustment directions is generated.

12. The method according to claim 1, characterized in that, The specific process for generating the structured earprint recognition and analysis report is as follows: Extract the original images of both auricles, the ear print texture type determination results, the quantitative confidence assessment results, and the abnormal prompt information generated during the acquisition process; Load the preset report template file, parse the data placeholders in the template, map and fill the extracted information into the corresponding placeholder positions in the template one by one, arrange the data blocks according to the preset content order, and generate a structured report file that conforms to the standard format. Write the report file to the specified storage path and push the report data to the front-end interface for rendering and display.

Images