Mandibular margin light and shadow feature extraction method and system based on visual reconstruction

Abstract

The present application relates to the technical field of feature extraction, in particular to a mandibular border light and shadow feature extraction method and system based on visual reconstruction, comprising the following steps: obtaining a mandibular border image and calculating the luminance parameters in blocks, classifying the task state according to the gray level and the change rate, extracting the gradient inversion boundary coordinates, screening the effective contour points, and superimposing to generate a light and shadow feature layer. In the present application, by establishing a data analysis mechanism of gray standard deviation and neighborhood luminance change rate at the image unit block level, the accurate expression of local light and shadow change characteristics can be realized, the classification and judgment of the task state are completed by combining the gray level change state and the spatial distribution characteristics, which is helpful to distinguish the real contour points between the mandibular border structure boundary and the light and shadow transition area, enhance the response ability to the real boundary of the mandibular border, improve the accuracy and stability of the boundary point set, and realize the unified reconstruction and accurate expression of the mandibular border light and shadow features based on the coordinate point set to construct the gray level change layer and the structure contour layer.

Description

Technical Field

[0001] This invention relates to the field of feature extraction technology, and in particular to a method and system for extracting mandibular border light and shadow features based on visual reconstruction. Background Technology

[0002] Feature extraction technology involves the process of identifying, extracting, and quantifying key visual elements from images, videos, or other visual data. This includes image edge detection, texture recognition, contour tracking, geometric modeling, and illumination and shadow information analysis. Typically, feature extraction and structural analysis of target regions are achieved through image preprocessing, feature point localization, gradient calculation, and spatial relationship modeling to support subsequent image recognition, object detection, or 3D reconstruction tasks. Traditional mandibular border light and shadow feature extraction methods identify the boundary or contour features of the mandibular border region in an image by analyzing the light and shadow distribution. These methods typically employ grayscale gradient extraction, shadow contour separation based on illumination changes, image segmentation techniques to extract the region, and the construction of an approximate contour model of the mandibular border structure using two-dimensional image sequences. Further enhancement of contrast is achieved through image filters or by utilizing brightness variation patterns in specific regions to determine structural boundaries, thereby extracting the light and shadow features of the mandibular border from medical images or facial images.

[0003] Traditional mandibular border light and shadow feature extraction mainly relies on the overall gray-level gradient trend of the image for contour recognition. It lacks detailed analysis of the brightness variation characteristics in local areas of the image, resulting in insufficient ability to identify boundaries and light and shadow transition areas. In actual operation, it cannot accurately distinguish the brightness blurring transition area between structural boundaries and adjacent shadow areas, causing the feature extraction results to be prone to contour breaks or artifact interference. At the same time, in the process of image segmentation and approximate model construction, it cannot effectively respond to the non-uniformity of local features, affecting the accurate representation of the real structural edges. In addition, the overall processing flow does not have a sufficient response mechanism to edge continuity and gray-level derivative change trends, which limits the integrity of feature point extraction and the stability of structural reconstruction. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides a method for extracting mandibular border light and shadow features based on visual reconstruction, comprising the following steps: S1: Obtain a two-dimensional image of the mandibular border region, divide it into image unit blocks according to the set pixel size, calculate the grayscale standard deviation and neighborhood brightness change rate data of each unit block, and generate a set of regional brightness parameters. S2: Based on the grayscale standard deviation and neighborhood brightness change rate data of each image unit block in the region brightness parameter group, and combined with the preset standard deviation interval boundary and change rate interval, the image unit blocks are classified into structure initialization processing state, light and shadow transition judgment state, boundary fitting start state, corner point extraction loading state and contour fusion push state, and a structure task priority classification set is generated. S3: Based on the image unit blocks in the boundary fitting start state of the structure task priority classification set, extract the gradient of change between consecutive pixels in the gray value sequence, locate the pixel point where the gray value derivative reverses direction, and simultaneously filter out duplicate coordinates and non-contour points to generate a set of light and shadow change boundary coordinates. S4: Based on the set of light and shadow change boundary coordinates, and combined with the grayscale change area of ​​the image unit block under the structure initialization processing state, perform a corresponding position overlap comparison operation on the coordinates of the two, perform boundary clarity screening and response offset distance limit judgment, and screen all coordinates to generate a task state contour response point set. S5: Draw a continuous boundary line layer based on the task state contour response point set, and construct a grayscale change layer by reading the brightness distribution value of the original image according to the coordinates in the layer. Then, stack the unified layer data in sequence to generate a mandibular edge light and shadow feature layer set.

[0005] As a further embodiment of the present invention, the preset standard deviation interval boundary includes a first standard deviation interval, a second standard deviation interval, and a third standard deviation interval; When classifying image unit blocks When the grayscale standard deviation is lower than the upper limit of the first standard deviation interval, the corresponding image unit block is determined to be in the structure initialization processing state; When the grayscale standard deviation is between the first standard deviation interval and the second standard deviation interval and the neighborhood brightness change rate is lower than the median of the change rate interval, the corresponding image unit block is determined to be the light and shadow transition determination state. When the grayscale standard deviation is above the second standard deviation range and the neighborhood brightness change rate is above the median of the change rate range, the corresponding image unit block is determined to be in the boundary fitting start state. When the grayscale standard deviation is within the third standard range and the neighborhood brightness change rate reverses direction, the corresponding image unit block is determined to be in the corner point extraction loading state. When the grayscale standard deviation and the neighborhood brightness change rate are both within the upper limit of their respective intervals and the state distribution of adjacent unit blocks is arranged in a continuous boundary, the corresponding image unit block is determined to be in the contour fusion push state.

[0006] As a further aspect of the present invention, when performing boundary sharpness screening and response offset distance limit judgment... After extracting coordinate pairs at the same location, the edge gradient of the corresponding points of the overlapping coordinates is calculated. Coordinate points that are lower than the set boundary clarity judgment threshold are removed. The Manhattan distance between the retained coordinate points and the corresponding points in the structure initialization state is calculated to determine whether it exceeds the set maximum allowable distance limit of response offset. If it exceeds the limit, the corresponding coordinate points are removed.

[0007] As a further embodiment of the present invention, the regional brightness parameter set includes grayscale standard deviation distribution records, neighborhood brightness change rate matrix, and image unit block spatial mapping information; the structural task priority classification set includes structural initialization processing unit classification, light and shadow transition judgment unit classification, boundary fitting start unit classification, corner point extraction loading unit classification, and contour fusion push unit classification; the light and shadow change boundary coordinate set includes boundary derivative inversion point coordinates, non-repeating boundary point coordinates, and contour path fitting point series; the task status contour response point set includes edge clarity verification points, response offset limit points, and boundary fine screening effective points; and the mandibular edge light and shadow feature layer set includes continuous boundary line layer, grayscale change distribution layer, and unified overlay synthesis layer.

[0008] As a further aspect of the present invention, the step of obtaining the regional brightness parameter group is as follows: S111: Obtain a two-dimensional image of the mandibular border region, divide the image into image unit blocks according to the preset pixel size, extract all pixels inside the image unit blocks, detect the red, green and blue channel values ​​of each pixel, calculate the corresponding gray values, and generate a set of image unit gray values. S112: Perform statistical operations based on the image unit gray value set, calculate the gray standard deviation and extract the gray mean of the unit block in the corresponding eight adjacent directions, and combine the average difference between the gray mean of the central unit block and the gray mean of the neighborhood to generate a neighborhood brightness change rate index set. S113: Based on the corresponding values ​​of each unit block in the neighborhood brightness change rate index set, construct the brightness feature vector of each unit block through combination mapping, arrange them in order according to the spatial position of the image unit blocks in the image matrix, and generate a regional brightness parameter group.

[0009] As a further aspect of the present invention, the step of obtaining the structured task priority classification set is as follows: S211: Obtain the brightness parameter group of the region, synchronously retrieve the gray standard deviation and the brightness change rate of the neighborhood for each image unit block, perform item-by-item interval comparison judgment based on the preset first standard deviation interval, second standard deviation interval, third standard interval and the median of the change rate interval, and write the unit block status identifier to generate a unit block status judgment identifier set. S212: Based on the unit block state determination identifier set, perform state encoding mapping according to the structure initialization processing state, light and shadow transition determination state, boundary fitting start state, and corner point extraction loading state. For the contour fusion push state, synchronously detect the continuity of adjacent unit block state arrangement, establish state association relationship, and obtain the unit block structure state mapping table. S213: Based on the unit block structure state mapping table, merge them sequentially according to each state, establish a one-to-one correspondence between the image unit block index and the state priority, and generate a structure task priority classification set.

[0010] As a further aspect of the present invention, the step of obtaining the set of coordinates of the light and shadow change boundary is as follows: S311: Based on the structure task priority classification set, filter the image unit blocks of the boundary fitting start state, traverse the pixel gray values ​​in each target unit block in row and column order, calculate the gray value difference between adjacent pixels in the spatial dimension, record the positive and negative change trend of the gray value difference between pixels, and generate a pixel gray gradient change sequence. S312: Based on the pixel grayscale gradient change sequence, locate the pixel index position where the positive and negative change directions in the sequence are reversed for the first time. Combined with the original image coordinates, mark the two-dimensional position of the corresponding pixel in the image space, remove duplicate coordinate items with the same row and column numbers, and generate a set of pixel coordinates with reversed direction. S313: Based on the direction-reversed pixel coordinate set, retrieve the neighborhood grayscale contrast change values ​​in the original image item by item, remove pixels with contrast amplitude less than the set grayscale jump recognition lower limit, and exclude linear feature points perpendicular to the edge direction of the unit block, retain key coordinate points with grayscale difference showing contour trend, and generate a set of light and shadow change boundary coordinates.

[0011] As a further aspect of the present invention, the step of obtaining the task state contour response point set is as follows: S411: Obtain the set of coordinates of the light and shadow change boundary and the set of coordinates of the gray-scale change area of ​​the image unit block under the structure initialization processing state. Perform a one-to-one correspondence search for the two types of coordinates according to the row and column numbers of the image, retain the points with completely matching coordinates and remove the coordinate values ​​that do not appear in the intersection, and generate a set of coordinates with overlapping boundaries. S412: Based on the set of boundary coincidence coordinates, obtain the edge gradient value of each coincidence point in the original image, remove the coordinate points whose corresponding edge gradient is lower than the set boundary clarity judgment threshold, retain only the valid coordinate index items that meet the gradient requirements, and generate a set of edge clear coordinates. S413: Based on the edge clear coordinate group and the coordinates of the gray-scale change area of ​​the image unit block under the structure initialization processing state, calculate the sum of the pixel distances of the corresponding coordinate pairs in the horizontal and vertical directions as the Manhattan distance value, compare it with the set maximum allowable distance limit of response offset, eliminate the coordinate points that exceed the limit, establish a unified index of the remaining valid points, and generate the task state contour response point set.

[0012] As a further aspect of the present invention, the step of obtaining the mandibular border light and shadow feature layer set is as follows: S511: Based on the two-dimensional coordinate data of the task status contour response point set and the spatial position identifier of the corresponding image unit block, a continuous boundary path is constructed by connecting adjacent points, and all path segments are mapped to the blank layer canvas, and a uniform path stroke weight and color identifier are assigned to generate a contour line layer. S512: For each coordinate point in the contour line layer, extract the brightness value of the corresponding pixel according to the row and column index in the original image, record the brightness change value sequence in the path order, draw the grayscale texture filling structure corresponding to the brightness in an independent layer, and use grayscale linear mapping to adjust the layer brightness gradient to generate a grayscale change layer. S513: Based on the contour line layer and the grayscale change layer, perform graphic composition processing according to the layer stacking order, calibrate the layer pixel alignment index and assign transparency weights, construct a complete image channel matrix structure, output a composite graphic including brightness information and edge path, and generate a mandibular edge light and shadow feature layer set.

[0013] A mandibular border light and shadow feature extraction system based on visual reconstruction, comprising: The image block parameter extraction module is used to perform S1: acquire a two-dimensional image of the mandibular border region, divide it into image unit blocks according to the set pixel size, calculate the grayscale standard deviation and neighborhood brightness change rate data of each unit block, and generate a set of regional brightness parameters. The priority state classification module is used to execute S2: based on the gray standard deviation and neighborhood brightness change rate data of each image unit block in the region brightness parameter group, and combined with the preset standard deviation interval boundary and change rate interval, the image unit block is classified into structure initialization processing state, light and shadow transition judgment state, boundary fitting start state, corner point extraction loading state and contour fusion push state, and generates a structure task priority classification set. The light and shadow boundary extraction module is used to execute S3: according to the image unit blocks in the boundary fitting start state in the structure task priority classification set, extract the gradient of change between consecutive pixels in the gray value sequence, locate the pixel position where the gray value derivative has a reversed direction, simultaneously filter out duplicate coordinates and non-contour points, and generate a set of light and shadow change boundary coordinates; The contour response screening module is used to execute S4: based on the set of light and shadow change boundary coordinates, combined with the gray change area of ​​the image unit block in the structure initialization processing state, the corresponding position overlap comparison operation of the two coordinates is performed, boundary clarity screening and response offset distance limit judgment are performed, and all coordinates are screened to generate the task state contour response point set. The light and shadow feature construction module is used to execute S5: draw a continuous boundary line layer based on the task state contour response point set, and construct a grayscale change layer by reading the brightness distribution value of the original image according to the coordinates in the layer, and then superimpose the unified layer data to generate the mandibular edge light and shadow feature layer set.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by establishing a data analysis mechanism for grayscale standard deviation and neighborhood brightness change rate at the image unit block level, the precise expression of local light and shadow change features can be achieved. Combining grayscale change state and spatial distribution features, the classification and judgment of task state can be completed, which helps to distinguish the real contour points between the mandibular border structural boundary and the light and shadow transition area. By locating key edge pixels through gradient change inversion features and simultaneously eliminating redundant and non-structural points, the responsiveness to the real boundary of the mandibular border is enhanced. Combining edge clarity screening and response offset distance judgment improves the accuracy and stability of the boundary point set. Based on the coordinate point set, a grayscale change layer and a structural contour layer are constructed to achieve unified reconstruction and precise expression of the light and shadow features of the mandibular border. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a flowchart of the process for obtaining the regional brightness parameter group in this invention; Figure 3 This is a flowchart of the process for obtaining the task priority classification set of the present invention; Figure 4 This is a flowchart illustrating the process of obtaining the boundary coordinate set of light and shadow changes in this invention. Figure 5 This is a flowchart of the process for obtaining the task status contour response point set in this invention; Figure 6 This is a flowchart of the process for obtaining the mandibular border light and shadow feature layer set in this invention; Figure 7This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides a method for extracting mandibular border light and shadow features based on visual reconstruction, comprising the following steps: S1: Acquire a two-dimensional image of the mandibular border region, divide it into image unit blocks according to the set pixel size, detect the red, green and blue channel values ​​of all pixels in each unit block, calculate the grayscale standard deviation and neighborhood brightness change rate data of each unit block, and generate a set of regional brightness parameters. S2: Based on the grayscale standard deviation and neighborhood brightness change rate data corresponding to each image unit block in the region brightness parameter group, compare and judge the grayscale standard deviation of the current unit block with the boundary of the preset standard deviation interval: The preset standard deviation interval boundaries include the first standard deviation interval, the second standard deviation interval, and the third standard deviation interval; When the grayscale standard deviation is lower than the upper limit of the first standard deviation interval, the corresponding image unit block is determined to be in the structure initialization processing state; When the grayscale standard deviation is between the first and second standard deviations and the neighborhood brightness change rate is lower than the median of the change rate range, the corresponding image unit block is determined to be in the light and shadow transition determination state. When the grayscale standard deviation is above the second standard deviation range and the neighborhood brightness change rate is above the median of the change rate range, the corresponding image unit block is determined to be in the boundary fitting start state. When the grayscale standard deviation is in the third standard range and the neighborhood brightness change rate reverses direction, the corresponding image unit block is determined to be in the corner point extraction loading state. When the grayscale standard deviation and the neighborhood brightness change rate are both within the upper limit of their respective intervals and the state distribution of adjacent unit blocks is arranged in a continuous boundary, the corresponding image unit block is determined to be in the contour fusion push state. After completing the classification and labeling of all image unit blocks, a structure task priority classification set is generated; S3: Based on the image unit blocks that are divided into boundary fitting start-up states in the structure task priority classification set, extract the gradient of change between consecutive pixels in the gray value sequence, locate the pixel position where the gray value derivative reverses direction, simultaneously filter out duplicate coordinates and non-contour points, and generate a set of light and shadow change boundary coordinates. S4: Based on the set of boundary coordinates of light and shadow changes, and combined with the gray-scale change area of ​​the image unit block under the structural initialization processing state, perform a corresponding position overlap comparison operation on the coordinates of the two to perform boundary sharpness screening and response offset distance limit judgment: After extracting coordinate pairs at the same location, calculate the edge gradient of the corresponding points of the overlapping coordinates, compare it with the set boundary clarity judgment threshold, and remove coordinate points that are below the threshold. Calculate the Manhattan distance between the retained coordinate points and the corresponding points in the structure initialization state, and determine whether it exceeds the set maximum allowable distance limit for response offset. If it exceeds the limit, remove the corresponding coordinate points. After completing the overall coordinate screening operation, generate the task state contour response point set. S5: Based on the coordinate information of the task state contour response point set and the corresponding image unit block position identifier, draw a continuous boundary line layer, and construct a grayscale change layer by reading the brightness distribution value of the original image according to the coordinate position in the layer. Then, stack them in sequence to construct unified layer data and generate a mandibular edge light and shadow feature layer set.

[0020] The regional brightness parameter set includes grayscale standard deviation distribution records, neighborhood brightness change rate matrix, and image unit block spatial mapping information. The structural task priority classification set includes structural initialization processing unit classification, light and shadow transition judgment unit classification, boundary fitting start unit classification, corner point extraction and loading unit classification, and contour fusion push unit classification. The light and shadow change boundary coordinate set includes boundary derivative inversion point coordinates, non-repeating boundary point coordinates, and contour path fitting point series. The task status contour response point set includes edge clarity verification points, response offset limit points, and boundary fine screening effective points. The mandibular edge light and shadow feature layer set includes continuous boundary line layer, grayscale change distribution layer, and unified overlay synthesis layer.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S111: Obtain a two-dimensional image of the mandibular border region, divide the image into image unit blocks according to the preset pixel size, extract all pixels inside the image unit blocks, detect the red, green and blue channel values ​​of each pixel, calculate the corresponding gray values, and generate a set of image unit gray values. Using an image acquisition terminal equipped with a high-resolution charge-coupled device (CCD), optical capture of the target mandibular border region was performed under a standard D65 light source environment to obtain raw two-dimensional image data with a resolution of 1920 x 1080 pixels. This two-dimensional image data was then loaded into the image processing memory buffer. Based on a set image unit block size standard of 16 x 16 pixels, the raw two-dimensional image was divided into equally spaced horizontal and vertical sections using a grid segmentation algorithm, generating a total of 8100 independent image unit block indices. Subsequently, for each image unit block, the data was extracted by traversing the block according to the raster scan order. The unit contains 256 pixels. The brightness values ​​of each pixel in the red, green, and blue channels are read. Using a weighted averaging algorithm based on human visual characteristics, the weights of the red channel value (×0.299), the green channel value (×0.587), and the blue channel value (×0.114) are multiplied, and the results are added and rounded down to obtain the single-channel gray value corresponding to the pixel. After traversing all pixels in the unit block, the 256 calculated gray values ​​are stored in order of their relative coordinates within the block to construct the image unit gray value set corresponding to the image unit block. For example, when the RGB values ​​of a pixel in the shadow area of ​​a mandible are 180 for the red channel, 160 for the green channel, and 140 for the blue channel, the values ​​are calculated as follows: 180 × 0.299 = 53.82, 160 × 0.587 = 93.92, and 140 × 0.114 = 15.96. The sum of these three values ​​is 163.7. After rounding, the grayscale value of the pixel is 163. This process is repeated until all unit blocks have been processed.

[0022] S112: Perform statistical operations based on the set of gray values ​​of image units, calculate the gray standard deviation and extract the gray mean of the unit block in the corresponding eight directions, and combine the average difference between the gray mean of the central unit block and the gray mean of the neighborhood to generate a neighborhood brightness change rate index set. According to the statistical analysis and processing logic, all gray values ​​in the image unit's gray value set are read. First, the arithmetic mean of the set is calculated as the gray value mean of the central unit block. Then, the squared difference between each gray value in the set and this mean is calculated. The sum of all squared differences is divided by the total number of pixels, and the square root of the quotient is taken to obtain the gray value standard deviation of the image unit block, which is used to quantify the dispersion of texture within the block. Simultaneously, based on the row and column index of the image unit block in the original image, its eight neighboring directions, namely the upper left, upper right, and upper left, are retrieved. For adjacent image units in the upper right, left, right, lower left, lower, and lower right directions, obtain the gray-scale mean of each of the eight adjacent units, and calculate the arithmetic mean of these eight means as the neighborhood gray-scale mean. Then, calculate the absolute value of the difference between the gray-scale mean of the central unit and the neighborhood gray-scale mean, and divide this absolute value by the neighborhood gray-scale mean (if the denominator is zero, perform non-zero perturbation processing) to obtain the neighborhood brightness change rate. Associate the calculated gray-scale standard deviation with the neighborhood brightness change rate by index to generate a neighborhood brightness change rate index set. For example, if the average gray level of a certain unit block is 150 and the calculated standard deviation of gray level is 12.5, and the average gray levels of its eight surrounding units are 145, 148, 152, 150, 149, 151, 147, and 155 respectively, then the average of the eight neighbors is 149.625. The absolute value of the difference between the central average of 150 and the neighborhood average of 149.625 is 0.375. Therefore, the brightness change rate is calculated as 0.375 divided by 149.625, which is approximately equal to 0.0025. This indicator reflects the degree of abruptness of the block relative to the background.

[0023] S113: Based on the corresponding values ​​of each unit block in the neighborhood brightness change rate index set, construct the brightness feature vector of each unit block through combination mapping, arrange them in order according to the spatial position of the image unit blocks in the image matrix, and generate the regional brightness parameter group. The algorithm iterates through the neighborhood brightness change rate index set. For each image unit block, it combines the three numerical parameters—the mean gray level, the standard deviation of gray level, and the neighborhood brightness change rate—according to a predetermined feature arrangement protocol to construct a brightness feature vector of length 3. Subsequently, based on the row-first or column-first spatial arrangement order of the image unit blocks in the original image matrix, the brightness feature vectors of all unit blocks are sequentially filled into the corresponding positions of the three-dimensional parameter matrix to establish a parameter mapping relationship across the entire image, thereby generating a regional brightness parameter set containing spatial topological information and local lighting attributes. For example, for the image unit block in the 5th row and 8th column of the image, its feature vector is [150, 12.5, 0.0025]. This vector is written into the (5, 8) coordinate position of the parameter set matrix to ensure that subsequent processing can directly use this lighting feature data through spatial indexing.

[0024] Please see Figure 3 The specific steps of S2 are as follows: S211: Obtain the region brightness parameter group, synchronously retrieve the grayscale standard deviation and the neighborhood brightness change rate for each image unit block, perform item-by-item interval comparison judgment based on the first standard deviation interval, the second standard deviation interval, the third standard interval and the median of the change rate interval, and write the unit block status identifier to generate a unit block status judgment identifier set. Read the data from the region brightness parameter group, and for each image unit block, extract its grayscale standard deviation value and neighborhood brightness change rate value; pre-set the first standard deviation interval to 0 to 10, the second standard deviation interval to 10 to 30, and the third standard deviation interval to 30 and above, while setting the median value of the change rate interval to 0.05; execute the item-by-item comparison logic: if the grayscale standard deviation of the unit block falls within the first standard deviation interval and the neighborhood brightness change rate is less than the median value of the change rate interval, it is determined to be a flat area, and a "structure initialization processing status" flag is written; if the grayscale standard deviation falls within the second standard deviation interval, it is determined to be a light and shadow transition area, and a "light and shadow transition determination status" flag is written; if the grayscale standard deviation falls within the third standard deviation interval or the neighborhood brightness change rate is greater than the median value of the change rate interval, it is determined to be an edge feature area, and a "boundary fitting start status" flag is written; if the grayscale standard deviation is extremely high and the change rate changes abruptly, a "corner point extraction loading status" flag is written; after traversing all unit blocks, generate a unit block status determination flag set containing the classification information of all unit blocks in the image. For example, for the cell block in the aforementioned feature vector with a grayscale standard deviation of 12.5 (between 10 and 30) and a change rate of 0.0025 (less than 0.05), it is determined and marked as "light and shadow transition determination state".

[0025] S212: Based on the unit block state determination identifier set, perform state encoding mapping according to the structure initialization processing state, light and shadow transition determination state, boundary fitting start state, and corner point extraction loading state. For the contour fusion push state, synchronously detect the continuity of adjacent unit block state arrangement, establish state association relationship, and obtain the unit block structure state mapping table. Based on the unit block state determination identifier set, the "structural initialization processing state" is first encoded as state code 01, the "light and shadow transition determination state" as state code 02, the "boundary fitting start state" as state code 03, and the "corner point extraction loading state" as state code 04. Then, the contour fusion push logic is started, and the consistency of the eight-neighborhood state is checked for each unit block. If a unit block in state code 02 has more than 4 unit blocks in its eight neighboring areas in state code 03, then according to the principle of state continuity, the state code of the central unit block is forcibly corrected to 03 to eliminate isolated noise points and connect broken edge features. Through this process, the state association relationship between unit blocks is established, the state codes are corrected and updated, and a unit block structure state mapping table with continuous spatial distribution and conforming to visual logic is obtained. For example, a unit block is initially determined to be in light and shadow transition (02), but its four adjacent blocks above, below, left, and right are all in boundary fitting state (03). The program executes the correction logic to update the state of the block to boundary fitting state (03) to ensure the integrity of the contour line.

[0026] S213: Based on the unit block structure state mapping table, merge sequentially according to each state, establish a one-to-one correspondence between the image unit block index and the state priority, and generate a structure task priority classification set. The system reads the unit block structure state mapping table and executes a priority merging strategy based on the importance of the status codes. "Corner Extraction Loading State" is defined as Level 1 priority, "Boundary Fitting Start State" as Level 2 priority, "Light and Shadow Transition Judgment State" as Level 3 priority, and "Structure Initialization Processing State" as Level 4 priority. The mapping table is traversed, establishing a one-to-one key-value pair relationship between the matrix index coordinates of each image unit block and its final determined state priority level. These key-value pairs are then rearranged in descending order of priority to generate a structure task priority classification set, ensuring that subsequent computing resources are preferentially allocated to high-frequency information areas such as corners and boundaries. For example, all boundary block indices marked as Level 2 priority are extracted and arranged before the light and shadow block indices marked as Level 3 priority, forming an ordered processing task queue.

[0027] Please see Figure 4 The specific steps of S3 are as follows: S311: Based on the structured task priority classification set, filter the image unit blocks of the boundary fitting start state, traverse the pixel gray values ​​in each target unit block in row and column order, calculate the gray value difference between adjacent pixels in the spatial dimension, record the positive and negative change trend of the gray value difference between pixels, and generate a pixel gray gradient change sequence. Based on the structural task priority classification set, all target image unit blocks marked as "boundary fitting initiation state" are selected. Within each target unit block, following a pixel scanning order from left to right and top to bottom, the gray values ​​of two adjacent pixels are read, and the difference between the gray values ​​of the next pixel and the previous pixel is calculated. If the difference is positive, the trend is recorded as "+"; if the difference is negative, the trend is recorded as "-"; if the difference is 0, the trend is recorded as "0". The difference trends of all adjacent pixel pairs within the block are connected in the scanning order to generate a pixel gray-level gradient change sequence reflecting local light and shadow fluctuations. For example, when scanning a row of pixels in an edge unit block, the gray-level value sequence is [50, 52, 55, 90, 88, 85], and the calculated difference is [+2, +3, +35, -2, -3]. The corresponding gradient change sequence is [+, +, +, -, -], where the transition from "+35" to "-2" reveals the potential edge location.

[0028] S312: Based on the pixel grayscale gradient change sequence, locate the pixel index position where the positive and negative change directions in the sequence are reversed for the first time. Combined with the original image coordinates, mark the two-dimensional position of the corresponding pixel in the image space, remove duplicate coordinate items with the same row and column numbers, and generate a set of pixel coordinates with reversed direction. The pixel grayscale gradient change sequence is analyzed to identify key nodes where the sign flips, specifically locating the peak inflection point where "+" changes to "-", or the valley inflection point where "-" changes to "+". The relative position of this inflection point within the cell block is obtained, and combined with the row and column offset of the cell block in the entire image, the absolute two-dimensional coordinates (X, Y) of the inflection pixel in the original image coordinate system are calculated. All identified inflection point coordinates are stored in a list, and a deduplication operation is performed to remove duplicate coordinate items with the same row and column numbers, ensuring that each physical point is recorded only once, generating a set of pixel coordinates with reversed direction. For example, in the above sequence, the sign changes from positive to negative at index 3, corresponding to the 4th position within the block. Combining the block's starting coordinates (100, 200), the absolute coordinates of this inflection point are calculated to be (103, 200), and this is added to the coordinate set.

[0029] S313: Based on the direction-reversed pixel coordinate set, retrieve the neighborhood grayscale contrast change values ​​in the original image item by item, remove pixels with contrast amplitude less than the set grayscale jump recognition lower limit, and at the same time exclude linear feature points perpendicular to the edge direction of the unit block, retain key coordinate points with grayscale difference showing contour trend, and generate a set of light and shadow change boundary coordinates. For each coordinate point in the direction-reversed pixel coordinate set, a 3x3 neighborhood is extracted from the original image centered on that point. The difference between the maximum and minimum grayscale values ​​within the neighborhood is calculated as the contrast change value. A grayscale jump detection lower limit of 30 is set. If the calculated contrast change value is less than 30, the point is considered a weak fluctuation caused by noise and is discarded. Simultaneously, the edge tangent direction of the point is calculated. If the linear feature formed by the point is detected to be perpendicular to the edge direction of the unit block at 90 degrees, it is determined to be a scanning artifact or non-contour feature and is excluded. Only key coordinate points that meet the contrast standard and conform to the contour extension trend are retained, generating a set of light and shadow change boundary coordinates. Table 1 below shows the data implementation reference for this screening process: Please refer to Table 1: Table 1. Key Coordinate Point Screening and Verification Data; As shown in Table 1, sample 001 has a contrast of 60, which is greater than the lower limit of 30, and the included angle of 15 degrees is not perpendicular, so it is retained; sample 002 has a qualified orientation but a contrast of only 20, so it is discarded; sample 003 has a high contrast but its features are perpendicular, so it is also discarded. The final output is a set of cleaned light and shadow change boundary coordinates.

[0030] Please see Figure 5 The specific steps of S4 are as follows: S411: Obtain the set of coordinates of the boundary of light and shadow changes and the set of coordinates of the gray-scale change area of ​​the image unit block under the structure initialization processing state. Perform a one-to-one correspondence search for the two types of coordinates according to the row and column numbers of the image, retain the points with completely matching coordinates and remove the coordinate values ​​that do not appear in the intersection, and generate the set of boundary coincident coordinates. The first dataset is the "set of light and shadow change boundary coordinates" generated above. The second dataset is the set of all pixel coordinates within the coverage area of ​​the image unit block marked as "structure initialization processing state" (i.e., flat background area) in S211. A set intersection operation is performed: for each coordinate point in the first dataset, it is checked whether it exists within the coordinate range included in the second dataset. If a coordinate point exists in both the light and shadow boundary set and falls within the area determined to be background (this usually occurs at the edge of a blurred shadow or in a misjudged area), it is considered a coincident point requiring further verification, and the coordinate is retained. Coordinates that are clearly located in non-coincident areas are removed, generating a set of coincident boundary coordinates. For example, if the coordinates (150, 208) are within the background area index range, they are included in the set of coincident boundary coordinates, indicating that the point may be located at the blurred boundary between the background and foreground.

[0031] S412: Based on the set of overlapping boundary coordinates, obtain the edge gradient value of each overlapping point in the original image, remove the coordinate points whose corresponding edge gradient is lower than the set boundary clarity judgment threshold, and retain only the valid coordinate index items that meet the gradient requirements to generate a set of edge clear coordinates. For each point in the set of coordinates coinciding at the boundary, its edge gradient magnitude is calculated on the original grayscale image. Specifically, the Sobel operator is used to calculate the horizontal gradient value Gx and the vertical gradient value Gy for that point. The absolute values ​​of Gx and Gy are added together to obtain the edge gradient value. A boundary sharpness threshold of 80 is pre-set. If the calculated edge gradient value is less than 80, the point is considered blurry and unreliable, and is discarded. Only strong edge points with gradient values ​​greater than or equal to 80 are retained, generating a set of sharp edge coordinates. For example, for coordinates (150, 208), if Gx is -50 and Gy is +40, the edge gradient value is |-50|+|40|=90, which is greater than the threshold of 80, and the point is retained. If the gradient sum of the other coordinates is 45, it is discarded.

[0032] S413: Based on the coordinates of the image unit block grayscale change area under the edge clear coordinate group and the structure initialization processing state, calculate the sum of the pixel distances of the corresponding coordinate pairs in the horizontal and vertical directions as the Manhattan distance value, compare it with the set maximum allowable distance limit of response offset, remove the coordinate points that exceed the limit, establish a unified index of the remaining valid points, and generate the task state contour response point set. The coordinates of the sharp edge coordinates are extracted as test points, and the geometric center coordinates of the image unit block under the "structure initialization processing state" are extracted as reference points. The Manhattan distance between the test point and the nearest reference point is calculated, which is the absolute value of the horizontal coordinate difference plus the absolute value of the vertical coordinate difference. The maximum allowable distance limit for response offset is set to 5 pixels. If the calculated Manhattan distance is greater than 5 pixels, it means that the edge point is too far from the background core area and is an invalid free point, so it is discarded. Otherwise, it is retained, and a unified index number is re-established for the remaining valid points to generate a task state contour response point set. For example, the horizontal difference between the test point (150, 208) and the nearest background center (152, 210) is 2, the vertical difference is 2, and the Manhattan distance is 4, which is less than 5. Therefore, this point is retained as a valid contour response point.

[0033] Please see Figure 6 The specific steps of S5 are as follows: S511: Based on the two-dimensional coordinate data of the task status contour response point set and the spatial position identifier of the corresponding image unit block, a continuous boundary path is constructed by connecting adjacent points, and all path segments are mapped to the blank layer canvas, giving a uniform path stroke weight and color identifier to generate a contour line layer. Read the 2D coordinate data of the task status contour response point set, and combine the spatial adjacency relationship of each point to its corresponding image unit block. Use spline interpolation algorithm to connect adjacent points to construct a smooth and continuous closed or semi-closed boundary path. Create a blank transparent layer canvas and map the constructed path to the canvas coordinate system. Set the path stroke weight to 2 pixels and the color identifier to pure black (RGB value 0, 0, 0). Draw a vector stroke along the mapped path to generate a contour line layer containing only geometric contour information. For example, connect the point set [(150, 208), (155, 212), (160, 215)] into a curve and draw a black line with a width of 2 pixels on the canvas to form the basic line drawing of the mandibular edge.

[0034] S512: For each coordinate point in the outline layer, extract the brightness value of the corresponding pixel according to the row and column index in the original image, record the brightness change value sequence in the path order, draw the grayscale texture fill structure corresponding to the brightness in an independent layer, and use grayscale linear mapping to adjust the layer brightness gradient to generate a grayscale change layer. Using the path coordinates in the outline layer as an index, the algorithm traces back to the original image to extract the brightness values ​​of each pixel at the corresponding coordinates and along its normal direction. Following the path sequence, a series of brightness change values ​​are recorded. A new independent layer is created, and based on this brightness sequence, a texture fill structure with corresponding grayscale values ​​is drawn around the path to simulate realistic light and shadow attenuation effects. A grayscale linear mapping algorithm is used to map the maximum value in the brightness sequence to 100% layer opacity and the minimum value to 0%. The alpha channels of the pixels within the layer are adjusted to generate a grayscale layer with dynamic opacity gradients. For example, if the brightness of a certain segment of the path gradually changes from 200 to 50 in the original image, a texture band that gradually changes from light gray to dark gray is drawn at the corresponding position in the grayscale layer to reflect the three-dimensionality of the jawline.

[0035] S513: Based on the contour line layer and grayscale change layer, the graphic composition process is performed according to the layer stacking order. The layer pixel alignment index is calibrated and the transparency weight is assigned. A complete image channel matrix structure is constructed, and a composite graphic including brightness information and edge path is output to generate a mandibular edge light and shadow feature layer set. Following the layer overlay logic, the bottom layer is set to a solid color background or the original image, the middle layer is a grayscale variation layer generated by S512, and the top layer is a contour line layer generated by S511. The pixel coordinate indices of each layer are aligned to ensure that the geometric contours and light and shadow textures are strictly matched in space. According to the preset visual weights, the opacity of the contour line layer is assigned to 80%, and the opacity of the grayscale variation layer is assigned to 60%. Alpha blending is performed to merge the layer pixel data and construct the final image channel matrix. This composite graphic, which integrates precise geometric edges and realistic light and shadow textures, is output, generating a mandibular edge light and shadow feature layer set. This layer set can be directly used for subsequent 3D modeling texture mapping or medical aesthetic analysis.

[0036] Please see Figure 7 A mandibular border light and shadow feature extraction system based on visual reconstruction, comprising: The image block parameter extraction module is used to perform S1: acquire a two-dimensional image of the mandibular border region, divide it into image unit blocks according to the set pixel size, calculate the grayscale standard deviation and neighborhood brightness change rate data of each unit block, and generate a set of regional brightness parameters. The priority state classification module is used to execute S2: Based on the gray standard deviation and neighborhood brightness change rate data of each image unit block in the region brightness parameter group, the image unit block is classified into structure initialization processing state, light and shadow transition judgment state, boundary fitting start state, corner point extraction loading state and contour fusion push state by combining the preset standard deviation interval boundary and change rate interval, and the structure task priority classification set is generated. The light and shadow boundary extraction module is used to execute S3: classify the image unit blocks of the set boundary fitting start state according to the priority of the structure task, extract the gradient of change between consecutive pixels in the gray value sequence, locate the pixel position where the gray value derivative has reversed direction, simultaneously filter out duplicate coordinates and non-contour points, and generate a set of light and shadow change boundary coordinates. The contour response screening module is used to execute S4: based on the set of boundary coordinates of light and shadow changes, combined with the gray-scale change area of ​​the image unit block under the structural initialization processing state, the corresponding position overlap comparison operation of the two coordinates is performed, boundary clarity screening and response offset distance limit judgment are performed, and all coordinates are screened to generate the task state contour response point set. The light and shadow feature construction module is used to execute S5: draw a continuous boundary line layer based on the task state contour response point set, and read the brightness distribution value of the original image according to the coordinates in the layer to construct a grayscale change layer. The unified layer data is then superimposed sequentially to generate the mandibular edge light and shadow feature layer set.

[0037] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A method for extracting mandibular border light and shadow features based on visual reconstruction, characterized in that, Includes the following steps: S1: Obtain a two-dimensional image of the mandibular border region, divide it into image unit blocks according to the set pixel size, calculate the grayscale standard deviation and neighborhood brightness change rate data of each unit block, and generate a set of regional brightness parameters. S2: Based on the grayscale standard deviation and neighborhood brightness change rate data of each image unit block in the region brightness parameter group, and combined with the preset standard deviation interval boundary and change rate interval, the image unit blocks are classified into structure initialization processing state, light and shadow transition judgment state, boundary fitting start state, corner point extraction loading state and contour fusion push state, and a structure task priority classification set is generated. S3: Based on the image unit blocks in the boundary fitting start state of the structure task priority classification set, extract the gradient of change between consecutive pixels in the gray value sequence, locate the pixel point where the gray value derivative reverses direction, and generate a set of light and shadow change boundary coordinates. S4: Based on the set of light and shadow change boundary coordinates, and combined with the grayscale change area of ​​the image unit block under the structure initialization processing state, perform a corresponding position overlap comparison operation on the coordinates of the two, perform boundary clarity screening and response offset distance limit judgment, and screen all coordinates to generate a task state contour response point set. S5: Draw a continuous boundary line layer based on the task state contour response point set, and construct a grayscale change layer by reading the brightness distribution value of the original image according to the coordinates in the layer. Then, stack the unified layer data in sequence to generate a mandibular edge light and shadow feature layer set.

2. The method for extracting mandibular border light and shadow features based on visual reconstruction according to claim 1, characterized in that: The preset standard deviation interval boundaries include a first standard deviation interval, a second standard deviation interval, and a third standard deviation interval; When the grayscale standard deviation is lower than the upper limit of the first standard deviation interval, the corresponding image unit block is determined to be in the structure initialization processing state; When the grayscale standard deviation is between the first standard deviation interval and the second standard deviation interval and the neighborhood brightness change rate is lower than the median of the change rate interval, the corresponding image unit block is determined to be the light and shadow transition determination state. When the grayscale standard deviation is above the second standard deviation range and the neighborhood brightness change rate is above the median of the change rate range, the corresponding image unit block is determined to be in the boundary fitting start state. When the grayscale standard deviation is within the third standard range and the neighborhood brightness change rate reverses direction, the corresponding image unit block is determined to be in the corner point extraction loading state. When the grayscale standard deviation and the neighborhood brightness change rate are both within the upper limit of their respective intervals and the state distribution of adjacent unit blocks is arranged in a continuous boundary, the corresponding image unit block is determined to be in the contour fusion push state.

3. The method for extracting mandibular border light and shadow features based on visual reconstruction according to claim 1, characterized in that: When performing boundary sharpness screening and response offset distance limit judgment... After extracting coordinate pairs at the same location, the edge gradient of the corresponding point is calculated. Coordinate points that are lower than the set boundary clarity judgment threshold are removed. The Manhattan distance between the retained coordinate points and the corresponding points in the structure initialization state is calculated to determine whether it exceeds the set maximum allowable distance limit for response offset. If it exceeds the limit, the corresponding coordinate point is removed.

4. The method for extracting mandibular border light and shadow features based on visual reconstruction according to claim 1, characterized in that: The region brightness parameter set includes grayscale standard deviation distribution records, neighborhood brightness change rate matrix, and image unit block spatial mapping information. The structure task priority classification set includes structure initialization processing unit classification, light and shadow transition judgment unit classification, boundary fitting start unit classification, corner point extraction and loading unit classification, and contour fusion push unit classification. The light and shadow change boundary coordinate set includes boundary derivative inversion point coordinates, non-repeating boundary point coordinates, and contour path fitting point series. The task status contour response point set includes edge clarity verification points, response offset limit points, and boundary fine screening effective points. The mandibular edge light and shadow feature layer set includes continuous boundary line layer, grayscale change distribution layer, and unified overlay synthesis layer.

5. The method for extracting mandibular border light and shadow features based on visual reconstruction according to claim 1, characterized in that, The steps for obtaining the region brightness parameter group are as follows: S111: Obtain a two-dimensional image of the mandibular border region, divide the image into image unit blocks according to the preset pixel size, extract all pixels inside the image unit blocks, detect the red, green and blue channel values ​​of each pixel, calculate the corresponding gray values, and generate a set of image unit gray values. S112: Perform statistical operations based on the image unit gray value set, calculate the gray standard deviation and extract the gray mean of the unit block in the corresponding eight adjacent directions, and combine the average difference between the gray mean of the central unit block and the gray mean of the neighborhood to generate a neighborhood brightness change rate index set. S113: Based on the corresponding values ​​of each unit block in the neighborhood brightness change rate index set, construct the brightness feature vector of each unit block through combination mapping, arrange them in order according to the spatial position of the image unit blocks in the image matrix, and generate a regional brightness parameter group.

6. The method for extracting mandibular border light and shadow features based on visual reconstruction according to claim 1, characterized in that, The steps for obtaining the structured task priority classification set are as follows: S211: Obtain the brightness parameter group of the region, synchronously retrieve the gray standard deviation and the brightness change rate of the neighborhood for each image unit block, perform item-by-item interval comparison judgment based on the preset first standard deviation interval, second standard deviation interval, third standard interval and the median of the change rate interval, and write the unit block status identifier to generate a unit block status judgment identifier set. S212: Based on the unit block state determination identifier set, perform state encoding mapping according to the structure initialization processing state, light and shadow transition determination state, boundary fitting start state, and corner point extraction loading state. For the contour fusion push state, synchronously detect the continuity of adjacent unit block state arrangement, establish state association relationship, and obtain the unit block structure state mapping table. S213: Based on the unit block structure state mapping table, merge them sequentially according to each state, establish a one-to-one correspondence between the image unit block index and the state priority, and generate a structure task priority classification set.

7. The method for extracting mandibular border light and shadow features based on visual reconstruction according to claim 1, characterized in that, The steps for obtaining the set of boundary coordinates for the light and shadow changes are as follows: S311: Based on the structure task priority classification set, filter the image unit blocks of the boundary fitting start state, traverse the pixel gray values ​​in each target unit block in row and column order, calculate the gray value difference between adjacent pixels in the spatial dimension, record the positive and negative change trend of the gray value difference between pixels, and generate a pixel gray gradient change sequence. S312: Based on the pixel grayscale gradient change sequence, locate the pixel index position where the positive and negative change directions in the sequence are reversed for the first time. Combined with the original image coordinates, mark the two-dimensional position of the corresponding pixel in the image space, remove duplicate coordinate items with the same row and column numbers, and generate a set of pixel coordinates with reversed direction. S313: Based on the direction-reversed pixel coordinate set, retrieve the neighborhood grayscale contrast change values ​​in the original image item by item, remove pixels with contrast amplitude less than the set grayscale jump recognition lower limit, and exclude linear feature points perpendicular to the edge direction of the unit block, retain key coordinate points with grayscale difference showing contour trend, and generate a set of light and shadow change boundary coordinates.

8. The method for extracting mandibular border light and shadow features based on visual reconstruction according to claim 1, characterized in that, The steps for obtaining the task status contour response point set are as follows: S411: Obtain the set of coordinates of the light and shadow change boundary and the set of coordinates of the gray-scale change area of ​​the image unit block under the structure initialization processing state. Perform a one-to-one correspondence search for the two types of coordinates according to the row and column numbers of the image, retain the points with completely matching coordinates and remove the coordinate values ​​that do not appear in the intersection, and generate a set of coordinates with overlapping boundaries. S412: Based on the set of boundary coincidence coordinates, obtain the edge gradient value of each coincidence point in the original image, remove the coordinate points whose corresponding edge gradient is lower than the set boundary clarity judgment threshold, retain only the valid coordinate index items that meet the gradient requirements, and generate a set of edge clear coordinates. S413: Based on the edge clear coordinate group and the coordinates of the gray-scale change area of ​​the image unit block under the structure initialization processing state, calculate the sum of the pixel distances of the corresponding coordinate pairs in the horizontal and vertical directions as the Manhattan distance value, compare it with the set maximum allowable distance limit of response offset, eliminate the coordinate points that exceed the limit, establish a unified index of the remaining valid points, and generate the task state contour response point set.

9. The method for extracting mandibular border light and shadow features based on visual reconstruction according to claim 1, characterized in that, The steps for obtaining the mandibular border light and shadow feature layer set are as follows: S511: Based on the two-dimensional coordinate data of the task status contour response point set and the spatial position identifier of the corresponding image unit block, a continuous boundary path is constructed by connecting adjacent points, and all path segments are mapped to the blank layer canvas, and a uniform path stroke weight and color identifier are assigned to generate a contour line layer. S512: For each coordinate point in the contour line layer, extract the brightness value of the corresponding pixel according to the row and column index in the original image, record the brightness change value sequence in the path order, draw the grayscale texture filling structure corresponding to the brightness in an independent layer, and use grayscale linear mapping to adjust the layer brightness gradient to generate a grayscale change layer. S513: Based on the contour line layer and the grayscale change layer, perform graphic composition processing according to the layer stacking order, calibrate the layer pixel alignment index and assign transparency weights, construct a complete image channel matrix structure, output a composite graphic including brightness information and edge path, and generate a mandibular edge light and shadow feature layer set.

10. A mandibular border light and shadow feature extraction system based on visual reconstruction, characterized in that, The system is used to implement the mandibular border light and shadow feature extraction method based on visual reconstruction as described in any one of claims 1-9, including: The image block parameter extraction module is used to perform S1: acquire a two-dimensional image of the mandibular border region, divide it into image unit blocks according to the set pixel size, calculate the grayscale standard deviation and neighborhood brightness change rate data of each unit block, and generate a set of regional brightness parameters. The priority state classification module is used to execute S2: based on the gray standard deviation and neighborhood brightness change rate data of each image unit block in the region brightness parameter group, and combined with the preset standard deviation interval boundary and change rate interval, the image unit block is classified into structure initialization processing state, light and shadow transition judgment state, boundary fitting start state, corner point extraction loading state and contour fusion push state, and generates a structure task priority classification set. The light and shadow boundary extraction module is used to execute S3: according to the image unit blocks in the boundary fitting start state in the structure task priority classification set, extract the gradient of change between consecutive pixels in the gray value sequence, locate the pixel position where the gray value derivative has a reversed direction, simultaneously filter out duplicate coordinates and non-contour points, and generate a set of light and shadow change boundary coordinates; The contour response screening module is used to execute S4: based on the set of light and shadow change boundary coordinates, combined with the gray change area of ​​the image unit block in the structure initialization processing state, the corresponding position overlap comparison operation of the two coordinates is performed, boundary clarity screening and response offset distance limit judgment are performed, and all coordinates are screened to generate the task state contour response point set. The light and shadow feature construction module is used to execute S5: draw a continuous boundary line layer based on the task state contour response point set, and construct a grayscale change layer by reading the brightness distribution value of the original image according to the coordinates in the layer, and then superimpose the unified layer data to generate the mandibular edge light and shadow feature layer set.

Images