Method and system for forensic blunt force trauma wound pattern feature image comparison analysis

Abstract

This invention belongs to the field of forensic image processing and pattern recognition technology, and discloses a method and system for image comparison and analysis of morphological features of forensic blunt force trauma wounds. The method includes: acquiring wound images using standardized macro photography with an orthophoto perspective and performing geometric correction; parameterizing the wound contour using an elliptical Fourier descriptor and extracting local geometric parameters; constructing a tissue retraction deformation compensation model based on elasticity mechanics for inverse correction; and matching the wounding tool cross-section with a cosine similarity database and outputting a recommended ranking of the wounding tools. This invention realizes the transformation of wound morphology from qualitative description to quantitative analysis.

Description

Technical Field

[0001] This invention belongs to the field of forensic image processing and pattern recognition technology, specifically involving a method and system for comparing and analyzing the morphological features of forensic blunt force trauma wounds. Background Technology

[0002] In forensic clinical and pathological examinations, lacerations caused by blunt force trauma to the human skin are among the most common types of injury. Statistics show that blunt force injuries account for over 60% of intentional injury cases, making it the most frequently encountered type of injury in forensic injury assessments. The morphological characteristics of lacerations, including the shape of the wound edges, the distribution of tissue bridges in the wound wall, and the bluntness or sharpness of the wound angles, have a close correspondence with the cross-sectional shape, size, and angle of impact of the injuring tool. This correspondence is crucial physical evidence for forensic diagnosis of the type of injuring object and reconstruction of the injury process, possessing irreplaceable evidentiary value for criminal case investigation and judicial trials. Accurately deducing the type and characteristics of the injuring object not only helps narrow down the scope of investigation and identify suspects but also provides scientific and objective technical support for court trials, possessing significant social value and practical significance.

[0003] However, traditional methods of identifying the causative agent primarily rely on forensic pathologists visually observing the wound's morphology and then recording its qualitative characteristics in writing, such as "irregular wound edges," "tissue bridging visible in the wound wall," or "one blunt and one sharp corner." This qualitative description method has inherent limitations: the same description can be applied to dozens of different causative agents, resulting in extremely low specificity of the inference and making it difficult to provide effective clues for narrowing down the scope of the investigation. Furthermore, different forensic pathologists may have significant differences in their written descriptions of the same wound, lacking objective and consistent evaluation standards, which restricts the repeatability and scientific rigor of the identification conclusions.

[0004] In the process of comparing and verifying the suspected weapon with the wound, current practices typically involve visually comparing the actual weapon with photographs of the wound, relying entirely on the forensic pathologist's spatial imagination and experience. However, after injury, skin tissue undergoes physiological responses such as elastic recoil and inflammatory swelling, resulting in significant deformation between the wound morphology at the time of image acquisition and the contact imprint of the weapon at the moment of injury. This deformation renders direct visual comparison unreliable. Especially in different anatomical locations with significant differences in tissue elasticity, the degree of recoil deformation varies considerably, further increasing the uncertainty of the comparison and judgment.

[0005] Patent application CN112053349A discloses a method for processing injury images for forensic identification. The technical solution includes acquiring the image to be identified, determining whether the image was taken vertically using a color segment scale, calculating the physical size parameters represented by each pixel, and calculating the length or area of ​​the wound based on these parameters. This solution solves the problem of accurately determining the specific size of wounds in images during remote forensic identification, enabling quantitative calculation of one-dimensional and two-dimensional dimensions such as the wound's length and area. However, this solution only focuses on wound size measurement and does not address the geometric parameterization of the wound's outline, nor does it provide an analytical tool for quantitative similarity matching between the wound's shape and the cross-sectional shape of the injuring tool. In other words, this solution can answer the question of how large the wound is, but it cannot answer the more investigatively valuable core question of what tool caused the wound.

[0006] Current forensic photography techniques merely record the appearance of wounds with ordinary cameras, lacking a complete analytical method for systematically describing the geometric parameters of wound morphology, considering tissue deformation compensation, and quantitatively matching it with the cross-sectional morphology of the injuring tool. Although some international scholars have attempted to use 3D scanning technology to conduct shape descriptive subclassification studies on simulated wounds, these methods rely on expensive 3D scanning equipment and primarily focus on wound classification rather than inferring the injuring tool, fundamentally differing from the technical problem addressed by this invention. Therefore, there is an urgent need for a comparative analysis method and system capable of geometrically parametrically extracting the wound contour from conventional 2D photographic images, simultaneously restoring the wound morphology to its instantaneous state using a tissue deformation compensation model, and quantitatively matching it with a database of injuring tool cross-sections. This would overcome the technical shortcomings of traditional qualitative description methods, such as low specificity and unreliable comparison verification relying on subjective judgment. Summary of the Invention

[0007] To address the technical problems in existing technologies where the description of the morphological characteristics of forensic blunt force trauma wounds relies on qualitative text, and the comparison and verification depend on subjective visual judgment without considering tissue retraction and deformation, this invention provides a method and system for image comparison and analysis of the morphological characteristics of forensic blunt force trauma wounds.

[0008] The forensic blunt force trauma wound morphological feature image comparison and analysis method provided by this invention includes the following steps:

[0009] Step S1, Standardized macro photography and wound image acquisition: Standardized macro photography is performed on skin lacerations caused by blunt force trauma from an orthophoto perspective. A geometric calibration target is placed around the wound. A high-resolution digital image containing the wound area and the geometric calibration target is acquired. The pixel physical resolution of the image is calculated based on the known geometric dimensions of the geometric calibration target. The wound area image is then geometrically corrected using the pixel physical resolution to eliminate lens distortion.

[0010] Step S2, wound contour extraction and Fourier descriptor parameterization: edge detection is performed on the geometrically corrected wound region image to extract the wound edge contour curve. The wound edge contour curve is parameterized and decomposed using an elliptical Fourier descriptor to obtain a set of frequency coefficients representing the overall shape of the wound. At the same time, the wound angle, wound edge curvature distribution and maximum width-to-length ratio of the wound are extracted as local geometric parameters.

[0011] Step S3, Tissue retraction deformation compensation correction: Based on the elastic mechanical properties of skin tissue, a tissue retraction deformation compensation model is constructed. According to the tissue elastic parameters corresponding to the anatomical location of the wound, the frequency coefficient set and local geometric parameters are inversely deformed and corrected to generate the deformed and compensated corrected Fourier descriptor and corrected local geometric parameters.

[0012] Step S4, Similarity Matching and Inference of Wounding Tool Sections: The profiles of each blunt instrument section in the wounding tool section library are parameterized and stored using the same elliptical Fourier descriptor. The cosine similarity between the Fourier descriptor and the Fourier descriptor of each tool in the section library is calculated, and the weighted Euclidean distance with corrected local geometric parameters is fused to obtain a comprehensive matching score. The recommended list of wounding tool types and their corresponding matching scores are output in descending order of the comprehensive matching score.

[0013] The above method uses Fourier descriptors to parametrically represent the wound contour, combines a tissue retraction deformation compensation model to restore the wound morphology to its state at the moment of injury, and then performs cosine similarity matching with a library of wound-causing tool sections. This achieves a transformation from qualitative description to quantitative analysis, providing objective and quantifiable reference data for forensic inference of the wound-causing agent. In this method, step S1 provides a geometrically corrected high-quality input image for step S2. The Fourier descriptors and local geometric parameters extracted in step S2 are fed into step S4 for matching calculation after deformation compensation correction in step S3. When the matching discrimination is insufficient, step S4 feeds back the candidate tool parameters to step S3 for iterative optimization. A deeply coupled closed-loop collaborative architecture of forward driving and backward feedback is formed between the steps, and the overall matching accuracy is significantly better than the effect of each step running alone.

[0014] This invention also provides a forensic blunt force trauma wound morphological feature image comparison and analysis system, including a standardized image acquisition module, a contour parameterization module, a deformation compensation module, and a similarity matching module. The standardized image acquisition module is used to acquire high-resolution images of the wound containing a geometrically calibrated target from an orthophoto perspective and perform geometric correction; the contour parameterization module is used to extract the wound edge contour and perform parametric decomposition using an elliptical Fourier descriptor; the deformation compensation module is used to perform inverse deformation correction based on the elastic parameters of the anatomical location; and the similarity matching module is used to calculate similarity with a library of injuring tool sections and output the matching ranking results. Each module corresponds one-to-one with the above method steps, working together to achieve the quantitative comparison and analysis function of wound morphological features.

[0015] Compared with existing technologies, this invention has the following beneficial effects: First, by using elliptic Fourier descriptors to geometrically parameterize the wound contour, it transforms traditional qualitative textual descriptions into calculable and comparable mathematical vectors, eliminating subjective differences between different observers and improving the objectivity and consistency of wound morphology descriptions. Second, by constructing a tissue retraction deformation compensation model based on elasticity mechanics, it effectively solves the interference problem of wound morphology deformation caused by tissue elastic retraction on matching, improving matching accuracy by approximately 16.7 percentage points compared to the uncompensated scheme. Third, through a dual-channel fusion matching degree scoring strategy and a closed-loop feedback correction mechanism, it achieves quantitative similarity matching between the wound contour and the cross-section of the injuring tool, providing a quantitative reference for forensic inference of the injuring agent. Attached Figure Description

[0016] Figure 1 This is a flowchart of the forensic blunt force trauma wound morphological feature image comparison and analysis method provided in the embodiments of the present invention;

[0017] Figure 2 This is an architecture diagram of the forensic blunt force trauma wound morphological feature image comparison and analysis system provided in this embodiment of the invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the following embodiments are only used to illustrate the technical concept of the present invention and are not intended to limit the scope of protection of the present invention. Various equivalent modifications and substitutions made by those skilled in the art after reading this specification without departing from the core idea of ​​the present invention all fall within the scope of protection of the claims of the present invention.

[0019] like Figure 1As shown, the forensic blunt force trauma wound morphological feature image comparison and analysis method provided by this invention includes steps S1 to S4, forming a closed-loop collaborative architecture of forward data flow driving and backward feedback correction among the steps. Specifically, the output of step S1 serves as the key input of step S2, and the output of step S2, after correction by step S3, is sent to step S4 for matching. When the matching discrimination is insufficient, step S4 can feed back the candidate tool parameters to step S3 for iterative optimization, thereby achieving adaptive closed-loop adjustment of the entire analysis process. The specific implementation methods of each step are described in detail below.

[0020] Step S1: Standardized macro photography and wound image acquisition. In one embodiment of the present invention, step S1 aims to acquire high-resolution digital images of the wound that meet the accuracy requirements of subsequent quantitative analysis. Considering the actual environmental constraints of forensic scene investigation and case examination, the present invention has strictly standardized the image acquisition process to ensure the comparability of images acquired by different operators and under different equipment conditions.

[0021] Specifically, the operator performs macro photography on skin lacerations caused by blunt force trauma from an orthophoto perspective. An orthophoto perspective means that the camera's optical axis is substantially aligned with the normal direction of the skin surface where the wound is located, with an angle between them not exceeding 5°. This angular constraint is crucial for subsequent geometric correction and contour extraction, as a large angle deviation will introduce perspective distortion, causing a non-negligible deviation between the projection of the wound contour onto the image plane and its actual shape. Preferably, the shooting distance is set to 15cm to 30cm, a range that ensures sufficient spatial resolution while adequately covering the wound area within the image field of view. In one embodiment of the invention, a digital camera equipped with a macro lens is used for shooting, with an image resolution of no less than 4000×3000 pixels, ensuring that the physical pixel resolution of the wound area reaches a precision level of 0.05mm / pixel or higher at the aforementioned shooting distance.

[0022] Before taking the picture, the operator needs to place a geometric calibration target around the wound. In one embodiment of the present invention, the geometric calibration target is a rigid thin sheet printed with a precise pattern. Its surface contains at least four non-collinearly distributed circular markers, each with a diameter of 3mm to 5mm. The center distance between adjacent markers has been pre-measured and recorded precisely. The function of the geometric calibration target is to provide a mapping reference between the image coordinate system and the physical coordinate system, enabling the system to accurately calculate the pixel physical resolution of the current image based on the number of pixels occupied by the markers with known geometric dimensions on the geometric calibration target. Preferably, the geometric calibration target also includes a set of checkerboard patterns for subsequent lens distortion parameter calibration.

[0023] After image acquisition, the system performs geometric correction on the wound area image using homography matrix transformation, based on the correspondence between the known physical coordinates of each marker point in the geometric calibration target and their pixel coordinates in the image. In one embodiment of the present invention, the geometric correction process includes two levels: the first level is lens distortion correction, which involves extracting the pixel coordinates of each corner point in the checkerboard pattern and estimating the radial distortion coefficient of the lens using the Zhang Zhengyou calibration method. , and tangential distortion coefficient , ,in The first-order radial distortion coefficient typically ranges from 1 to 10. to ; These are the second-order radial distortion coefficients, and their values ​​typically range from [value range missing]. to ; and The tangential distortion coefficient typically ranges from 1 to 10. to Then, the original image is subjected to anti-distortion mapping using the aforementioned distortion coefficients. The second level is perspective correction, which calculates the homography matrix based on the coordinate correspondence of four or more marker points. The image is then transformed to an orthographic projection plane to eliminate perspective distortion caused by minor deviations in the shooting angle. After these two levels of geometric correction, the corrected image of the wound area and its corresponding pixel physical resolution are output. The unit is mm / pixel. The pixel physical resolution... The calculation formula is: ,in: The known physical distance between two marked points on the geometric calibration target, in mm, is determined by the factory parameters of the geometric calibration target, with a typical value of 20 mm to 50 mm. The distance between the two marker points in the corrected image is expressed in pixels, obtained through a marker point center detection algorithm. This formula establishes a linear mapping between pixel space and physical space, providing a unified metric for subsequent contour extraction and parameter calculation.

[0024] Preferably, after geometric correction and before proceeding to step S2, the present invention further preprocesses the corrected image to enhance the contrast between the wound area and the surrounding normal skin. The preprocessing includes the following three sub-steps: First, the original RGB color image is converted to the Lab color space, and the a-channel image is extracted. Since lacerations in forensic medicine are typically dark red or purplish-red, while normal skin is skin-colored, the a-channel (red-green component) can effectively distinguish the wound area from the background skin. Second, a bilateral filter is applied to the a-channel image for denoising. The spatial domain standard deviation of the bilateral filter is set to 10 pixels, and the value domain standard deviation is set to 50 gray levels. This filter effectively preserves the gray-level transitions at the wound edge while smoothing noise, avoiding false edges in subsequent edge detection due to noise. Finally, contrast-limited adaptive histogram equalization (CLAHE) is used to locally enhance the contrast of the filtered a-channel image. The block size of CLAHE is set to 8×8 pixels, and the contrast limiting parameter is set to 3.0. This processing can locally enhance the edge gradient in the wound area, which is beneficial to the accuracy of subsequent multi-scale Canny edge detection. The parameters of the above preprocessing process were selected through experiments on more than 50 images with different lighting conditions and different wound types. The noise false detection rate was controlled below 5% while ensuring the wound edge enhancement effect.

[0025] Step S2: Wound contour extraction and Fourier descriptor parameterization. Step S2 takes the corrected image output from step S1 as input to complete the extraction and parameterization of the wound edge contour. This is the core step of the present invention to achieve quantitative description of wound morphology.

[0026] In a preferred embodiment of the present invention, the extraction of the wound edge contour employs a multi-scale Canny edge detection strategy. The reason for using a multi-scale approach instead of a single-scale edge detection is that the wound edges of lacerations are typically irregularly serrated, and false detections of edge breaks or noisy edges are easily encountered at a single Gaussian kernel scale. Specifically, the Gaussian kernel standard deviation is used to... , and Canny edge detection was performed on the wound area image at these three scales, among which... Corresponding to a fine scale, it can capture the micro-serrated features of the wound edge; Corresponding to intermediate scales, it balances edge continuity with detail preservation; For the coarse scale, the overall contour of the wound is mainly extracted. The high and low thresholds for Canny detection at each scale are set as follows: , ( scale), , ( scale), , ( (scale), among which and These represent the low and high thresholds in the Canny edge detection algorithm, expressed in grayscale levels (0 to 255). The low threshold is used for edge extension tracking, while the high threshold is used for strong edge detection. Subsequently, the edge detection results from the three scales are logically ORed and merged to obtain a preliminary edge map fusing multi-scale information. Further, a circular structuring element with a radius of 3 pixels is used to perform morphological closure operations on the fused edge map to eliminate edge breaks caused by wound tissue bridging or bloodstain interference. Finally, the connected component with the largest area is extracted as the wound edge contour curve.

[0027] Preferably, when the automatic edge detection results are unsatisfactory, the present invention also provides an interactive contour correction function. The operator can locally correct the automatically extracted contour curve on the display interface, manually adding or deleting some edge segments to ensure that the final contour curve accurately reflects the true edge direction of the wound. It should be noted that interactive correction is not a necessary step. In most typical laceration cases, the multi-scale Canny edge detection strategy combined with morphological closure operations can automatically obtain contour curves that meet the requirements of subsequent analysis. Interactive correction is only enabled when the wound edge is covered by a large amount of blood, has prominent tissue bridges, or the lighting is uneven, resulting in poor automatic extraction quality. In one embodiment of the present invention, in the above-mentioned 30 test samples, 26 cases (86.7%) obtained qualified contour curves without interactive correction.

[0028] After obtaining the wound edge contour curve, this invention uses elliptic Fourier descriptors (EFD) to parametrically decompose it. The elliptic Fourier descriptor is a mathematical tool that decomposes a closed planar curve into a set of harmonic components, each of which approximates the shape characteristics of the original curve through elliptical motion at different frequencies. Compared to the simple Fourier descriptor, the elliptic Fourier descriptor can simultaneously process the curve's... Components and The component exhibits good adaptability to both open and closed curves.

[0029] Specifically, let the contour curve of the wound edge be formed by... Equally spaced sampling points describe, ,in This represents the total number of contour sampling points, typically ranging from 200 to 500 points. The contour curve... coordinates and The coordinates are expanded into Fourier series, the first... The elliptic Fourier coefficients corresponding to the first harmonic are , , , The calculation formula is as follows:

[0030] .

[0031] .

[0032] .

[0033] .

[0034] in: The harmonic order is... , The maximum harmonic order is between 10 and 30, and the present invention preferably uses 20. This order can filter out high-frequency noise caused by tissue bridges and bloodstains while preserving the overall shape characteristics of the wound. This is the total arc length parameter of the contour curve, in pixels; To sample from the first sampling point along the contour to the second sampling point... The cumulative arc length of each sampling point ; For adjacent sampling points Coordinate difference, in pixels; For adjacent sampling points Coordinate difference, in pixels; This represents the arc length increment between adjacent sampling points, in pixels.

[0035] Through the above calculations, we obtain the result from... The set of frequency coefficients consisting of first harmonic coefficients To eliminate the influence of translation, rotation, and scaling of the contour curve on the descriptor, this invention further normalizes the set of frequency coefficients: using the semi-major axis length of the first harmonic as the scale normalization factor, and using the major axis direction of the first harmonic ellipse as the rotation alignment reference, the normalized Fourier descriptor vector is obtained. This normalization process is particularly important in wound morphology comparison analysis: in actual forensic work, wound images taken by different operators may differ in position, orientation, and scaling, while tissue retraction and swelling can also alter the absolute size of the wound. Through scale normalization and rotation alignment, this invention ensures that the descriptor reflects only the pure shape characteristics of the wound, unaffected by the aforementioned external factors, resulting in good consistency and repeatability of Fourier descriptors for the same wound collected at different times and under different conditions.

[0036] Simultaneously, this invention also extracts three local geometric parameters as supplementary features of the Fourier descriptor. The first is the creation angle. The first angle is defined as the angle between the tangents at both ends of the wound, ranging from 0° to 180°. The calculation method for the wound angle is as follows: at each end of the wound contour, take a 5mm long curve segment along the direction of contour advancement, and fit its tangent direction using the least squares method. The angle between the two tangent directions is the wound angle. The second is the wound edge curvature distribution. This represents the local curvature value at each point along the contour of the wound edge. For the arc length parameter, the curvature is calculated using the three-point approximation method, where the reciprocal of the radius of the circumcircle determined by three adjacent sampling points is used as the curvature value at the midpoint, with units of 1 / mm. The curvature value of the entire profile is quantized into a curvature histogram containing 20 bins. The box size range is 0 to 2.0 mm. The third is the maximum width-to-length ratio. Defined as the maximum transverse diameter of the wound With the maximum longitudinal diameter ratio The maximum transverse diameter and maximum longitudinal diameter are the width and height of the smallest bounding rectangle of the wound outline, respectively, in mm.

[0037] Step S3: Tissue Retraction Deformation Compensation and Correction. Step S3 is the key innovative step that distinguishes this invention from existing technologies. In forensic practice, after skin tissue is subjected to blunt force trauma resulting in a laceration, the wound edges retract due to the inherent elasticity of the skin. This leads to a significant difference between the wound's morphology at the moment of image acquisition and the morphology of the contact mark left by the injuring tool at the instant of injury. Furthermore, the inflammatory swelling response of the local tissue after injury further alters the wound's morphology. If this tissue deformation is not compensated for, directly comparing the acquired wound contour with the cross-section of the injuring tool will inevitably lead to increased matching errors and reduced reliability of the matching results. This invention constructs a tissue retraction deformation compensation model based on elasticity mechanics, reversibly restoring the wound contour from the "current observation state" to the "state at the moment of injury," thereby providing more accurate input data for subsequent similarity matching.

[0038] In one embodiment of the present invention, the tissue retraction deformation compensation model is constructed based on the following physical assumptions: skin tissue can be approximated as an isotropic linear elastic material within a small deformation range, and its mechanical behavior can be determined by the elastic modulus. Compared to Poisson Two parameters describe this. However, skin tissue in different anatomical locations varies in thickness, collagen fiber orientation, and subcutaneous fat content, resulting in elastic modulus that can differ by several times between different locations. Therefore, this invention establishes an elastic parameter lookup table for anatomical locations, dividing common injury sites in the human body into six regions: head and face, neck, chest and back, abdomen, upper limbs, and lower limbs. Based on experimental data reported in biomechanical literature, corresponding reference values ​​for elastic modulus and Poisson's ratio are assigned to each region.

[0039] Specifically, the elastic modulus The value ranges from 0.1 MPa to 10 MPa, with typical values ​​for the elastic modulus of the skin on the head and face being 0.5 MPa to 2.0 MPa, the neck 0.3 MPa to 1.5 MPa, the chest and back 0.8 MPa to 3.0 MPa, the abdomen 0.2 MPa to 1.0 MPa, the upper limbs 1.0 MPa to 5.0 MPa, and the lower limbs 1.5 MPa to 8.0 MPa. The Poisson's ratio is... For skin tissue, the value ranges from 0.3 to 0.5; the preferred value in this invention is 0.45, which reflects the near-incompressible mechanical properties of skin tissue. When performing step S3, the operator needs to specify the anatomical location of the wound, and the system automatically retrieves the corresponding elastic modulus from the lookup table. Compared to Poisson .

[0040] After obtaining the tissue elasticity parameters, this invention constructs a deformation field using the Radial Basis Function (RBF) method to perform inverse deformation correction on the wound contour. The basic principle is as follows: each sampling point on the wound edge contour curve is considered a point mass in an elastic body. Based on the equilibrium equations of elasticity, the radial displacement of each point mass under tissue retraction is estimated. Then, the contour curve is radially expanded by this displacement to restore the original shape of the wound at the moment of injury.

[0041] In one embodiment of the present invention, the radial retraction amount The calculation formula is: ,in: The arc length parameter of the wound contour is Radial retraction at the location, in mm; The empirical correction coefficient is dimensionless and ranges from 0.5 to 2.0. The preferred value in this invention is 1.2. This coefficient is used to compensate for the deviation between the theoretical model and the actual organizational behavior. Its value is based on the fitting analysis results of retrospective case data of 30 known injury-causing tools. Poisson's ratio is dimensionless. Elastic modulus, in MPa; This is an estimated value of the contact stress at the moment of injury, in MPa. This value is estimated based on the empirical relationship between wound depth and the impact force of the injury-causing object, and ranges from 0.01 MPa to 1.0 MPa, with 0.1 MPa being the preferred value. The arc length parameter of the wound contour is The distance from the point to the geometric center of the wound, in mm, reflects the spatial distribution characteristics of radial retraction at each point on the contour.

[0042] The physical meaning of the above formula is: elastic modulus The smaller (softer the tissue), the greater the shrinkage. The larger; Poisson's ratio The closer the value is to 0.5 (the closer the tissue is to incompressibility), the greater the shrinkage; while... The introduction of this factor results in a greater amount of retraction of the contour points farther away from the center of the wound, which is consistent with the actual observed wound retraction pattern, that is, the wound edge retracts towards the center.

[0043] The radial retraction amount at each sampling point was calculated. Then, each sampling point on the wound edge contour curve is expanded outward along its radial direction (i.e., the direction from the geometric center of the wound to that point). The distance is calculated to obtain the deformed profile curve. Then, the deformed profile curve is re-decomposed using elliptic Fourier descriptors to obtain the corrected Fourier descriptor. At the same time, the corrected wound angle is recalculated. Correction curvature histogram and corrected aspect ratio .

[0044] Preferably, the present invention also introduces a time decay factor. The amount of retraction is adjusted to account for differences in tissue deformation at different time intervals after injury. The formula for calculating the time decay factor is as follows: ,in: This is the time decay factor, which is dimensionless and ranges from 0 to 1. The time interval from the moment of injury to the moment of image acquisition, in hours (h). is a time constant in hours (h), reflecting the characteristic time required for tissue recoil to reach steady state. Its value ranges from 0.5h to 4.0h, and the preferred value in this invention is 2.0h. This value is based on data from skin tissue stress relaxation experiments, which show that most elastic recoil is completed within 2 hours after injury. Much larger hour, When the value approaches 1, it indicates that the pullback is basically complete. At this point, the pullback amount is equal to the value in the formula. ;when When approaching 0, A value approaching 0 indicates that shrinkage has not yet occurred, and deformation compensation is not required at this point. In practical application, the formula should be adjusted accordingly. Multiply As the final radial retraction amount.

[0045] Step S4: Similarity matching and inference of the wounding tool cross-section. Step S4 takes the corrected Fourier descriptor and corrected local geometric parameters output from Step S3 as input, and performs quantitative matching with the pre-established wounding tool cross-section library, outputting the most likely wounding tool type recommendation and matching degree score. This step is the core link of this invention in transforming wound morphology information into wounding tool inference conclusions.

[0046] In one embodiment of the present invention, the wounding tool cross-section library is constructed as follows: Common blunt force traumatizing objects used in forensic practice are collected, including at least four major categories: hammer face cross-sections, wrench cross-sections, tubular object cross-sections, and brick corner cross-sections. Each major category is further subdivided into different specifications and sizes. For example, hammer face cross-sections include circular hammer face (diameters of 20mm, 25mm, 30mm, 35mm, 40mm, etc., more than five specifications), square hammer face (side lengths of 20mm×20mm, 25mm×25mm, 30mm×30mm, etc.), and octagonal hammer face, etc.; wrench cross-sections include open-end wrenches (opening widths of 10mm, 12mm, 14mm, 17mm, 19mm, etc.) and box wrenches, etc.; tubular object cross-sections include round tubes (outer diameters of 15mm, 20mm, 25mm, 32mm, etc.) and square tubes, etc.; brick corner cross-sections include standard brick corners (90° right angles) and worn brick corners, etc. In one embodiment of the present invention, each type of blunt instrument cross-sectional profile includes no less than 5 cross-sectional samples of different sizes, and the total number of samples in the entire cross-sectional library is no less than 40.

[0047] The establishment of the cross-section library has strict standardization requirements. Specifically, the method for obtaining each tool cross-section sample is as follows: the striking surface of the actual blunt object is placed vertically on a flatbed scanner, and a digital image of the cross-section is obtained at a scanning resolution of 600 dpi. Then, the cross-section contour curve is extracted using the same edge detection and contour extraction algorithm as in step S2. For irregularly shaped tools (such as brick corners), an industrial digital camera is used in conjunction with a geometrically calibrated target to take cross-section photos from an orthophoto perspective before extracting the contour. To ensure the completeness and representativeness of the cross-section library, this invention suggests that the construction of the cross-section library refer to the classification catalog of common injury-causing substances in the criminal technical standards of the Ministry of Public Security, and that the cross-section samples be regularly supplemented and updated according to new injury-causing substances appearing in actual cases. Preferably, the cross-section library is stored using a hierarchical index structure. The first layer is classified according to the major categories of injury-causing substances (hammers, wrenches, pipes, bricks, etc.), the second layer is classified according to specific shapes (circular, square, hexagonal, etc.), and the third layer is arranged according to specifications and dimensions, which facilitates rapid retrieval and matching range filtering.

[0048] For each tool cross-section sample in the cross-section library, its cross-section profile is parametrically decomposed and normalized using the same elliptic Fourier descriptor method as in step S2 to obtain the Fourier descriptor vector of the tool side. subscript Indicates the first section in the section library Each tool sample is analyzed. Local geometric parameters such as angle, curvature histogram, and width-to-length ratio of each tool's cross-section are extracted simultaneously.

[0049] In the matching calculation stage, this invention employs a dual-channel fusion matching score strategy. The first channel calculates the global shape similarity, using cosine similarity as a metric to correct the Fourier descriptor. With tool descriptors The degree of similarity between them. After flattening the Fourier coefficients into a one-dimensional vector, the formula for calculating cosine similarity is: ,in: For wound correction descriptor and the first The cosine similarity between the tool descriptors, with values ​​ranging from 1 to 2. to The closer the value is It indicates that the more similar the shapes are; To correct the one-dimensional vector after flattening the Fourier descriptor, the dimension is... ,Right now ; For the first Each tool descriptor is a flattened one-dimensional vector, also with dimension 1. ; This represents the vector dot product operation; This represents the Euclidean norm of a vector.

[0050] The second channel calculates the distance between local geometric parameters, using a weighted Euclidean distance metric to correct for discrepancies between the local geometric parameters and the tool's local geometric parameters. The formula for calculating the weighted Euclidean distance is: ,in: To correct local parameters of the wound and the first The weighted Euclidean distance between the local parameters of each tool is dimensionless. , , These are the weighting coefficients for the three feature components: the creation angle, the curvature histogram, and the width-to-length ratio. They are dimensionless, and the preferred values ​​in this invention are... , , The weighting is based on the fact that the angle of injury contributes the most to distinguishing the type of injuring tool, followed by the curvature distribution, and the width-to-length ratio is used as an auxiliary verification. and These are respectively the wound correction angle and the first The cross-sectional angle of each tool, in degrees, is normalized by dividing by 180. and These are the wound correction curvature histogram and the first... The vector of the curvature histogram of each tool after L2 norm normalization; and These are the corrected width-to-length ratio of the wound and the first The aspect ratio of each tool; This is the maximum width-to-length ratio of all tools in the section library, used for normalization; typical values ​​range from 3.0 to 5.0.

[0051] The formula for calculating the overall match score is: ,in: For the wound and the first The overall matching score of each tool, with a value ranging from 0 to 1, where a higher value indicates a higher degree of matching; The first weight coefficient is dimensionless and ranges from 0.6 to 0.8. The preferred value in this invention is 0.7. This weight makes the global shape similarity dominate the comprehensive score, reflecting the core role of the Fourier descriptor in overall shape matching. This is the second weighting coefficient, with a value ranging from 0.2 to 0.4; The distance attenuation coefficient is dimensionless and ranges from 1.0 to 5.0. The preferred value in this invention is 2.5. This coefficient controls the attenuation rate of the local parameter distance on the matching score. The larger the value, the stronger the penalty for the score. This is a negative exponential mapping value for the distance between local parameters. It takes the value of 1 when the local parameters are completely identical, and the larger the distance, the closer the value is to 0.

[0052] After the calculation is completed, the system will give a comprehensive matching score to all tool samples. Sort in descending order, output a recommended list of wound-causing tool types and their corresponding matching scores. Preferably, the system outputs the top 5 matching results and displays a visual comparison of the cross-sectional profile of the top-ranked tool and the corrected wound profile.

[0053] In a preferred embodiment of the present invention, step S4 further includes a feedback correction step to improve analysis accuracy when the discrimination of the matching results is insufficient. When the comprehensive matching score of the top-ranked injury tool type... Overall match score ranked second difference Less than the preset discrimination threshold When this occurs, feedback correction is triggered. The discrimination threshold... The value ranges from 0.02 to 0.10, and the preferred value in this invention is 0.05. The specific method of feedback correction is as follows: the known precise geometric parameters of the first-ranked damaging tool's cross-sectional profile are used as constraints and input back into the tissue retraction deformation compensation model in step S3, adjusting the empirical correction coefficients accordingly. and contact stress estimates Iterative optimization is performed. The objective function is to minimize the Hausdorff distance between the corrected wound contour and the top-ranked tool section contour. Gradient descent is used for iterative optimization with a learning rate of 0.01, a maximum of 50 iterations, and a convergence criterion of less than 0.001 for the change in the overall matching score between two consecutive iterations. After optimization, steps S3 and S4 are re-executed with updated compensation parameters, and the final output is the matching result after feedback correction. This closed-loop feedback mechanism effectively improves matching accuracy even in edge cases where discrimination is insufficient, demonstrating the deep coupling and synergistic relationship between steps S3 and S4.

[0054] like Figure 2 As shown, this invention also provides a forensic blunt force trauma wound morphological feature image comparison and analysis system. This system includes four functional modules: a standardized image acquisition module, a contour parameterization module, a deformation compensation module, and a similarity matching module. Each module corresponds one-to-one with steps S1 to S4 in the above method embodiments, working together to achieve the quantitative comparison and analysis function of wound morphological features. The specific implementation methods of each module are described below.

[0055] The standardized image acquisition module corresponds to step S1 in the method embodiment, and is configured to perform standardized macro photography of skin lacerations caused by blunt force trauma from an orthophoto perspective. At the hardware level, this module includes a digital camera equipped with a macro lens or a smart terminal device with macro shooting capabilities, and a geometric calibration target placed around the wound. At the software level, the module includes an image acquisition control subunit and a geometric correction subunit. The image acquisition control subunit controls shooting parameters, including exposure time, aperture value, ISO sensitivity, and white balance, to ensure consistent image quality under different ambient lighting conditions. The geometric correction subunit performs geometric correction on the wound area image according to the two-level correction method (lens distortion correction and perspective correction) described in the method embodiment, based on the correspondence between the known physical coordinates of each marker point on the geometric calibration target and its pixel coordinates in the image. Preferably, the geometric correction subunit uses the `calibrateCamera` and `warpPerspective` functions from the OpenCV library to implement the above correction algorithm, with a correction processing time of no more than 0.5 seconds. The output of the standardized image acquisition module is a geometrically corrected image of the wound area and its corresponding pixel physical resolution. The output data is stored in lossless PNG format and is accompanied by a JSON format correction parameter metadata file.

[0056] The contour parameterization module corresponds to step S2 in the method embodiment. It is configured to perform edge detection on the geometrically corrected wound region image to extract the wound edge contour curve, and then perform parameterized decomposition using an elliptical Fourier descriptor. This module includes a multi-scale edge detection subunit, a contour interaction correction subunit, and a Fourier parameterization subunit. The multi-scale edge detection subunit implements the three scales described in the method embodiment (…). , , The system performs Canny edge detection, logical OR merging, and morphological closure operations. The contour interactive correction subunit provides a graphical user interface, allowing operators to locally correct automatically extracted contour curves on the screen. It supports interactive operations such as dragging contour control points, adding new control points, and deleting erroneous edge segments. The Fourier parameterization subunit performs parametric decomposition and normalization of the contour curves according to the elliptic Fourier descriptor formula described in the method embodiments. Furthermore, this subunit simultaneously extracts three local geometric parameters: the creation angle, the creation edge curvature histogram, and the maximum width-to-length ratio. In one embodiment of the invention, the entire calculation process of the contour parameterization module takes no more than 2 seconds to run on a general-purpose computer equipped with an Intel Core i5 or higher processor and 8GB or more of memory.

[0057] The deformation compensation module corresponds to step S3 in the method embodiment. It is configured to perform inverse deformation correction on the frequency coefficient set and local geometric parameters output by the contour parameterization module based on the elastic mechanical properties of skin tissue and the tissue elastic parameters corresponding to the anatomical location of the wound. This module includes an anatomical location selection subunit, an elastic parameter lookup subunit, and a radial deformation calculation subunit. The anatomical location selection subunit provides a human contour diagram interface. The operator can select the anatomical location of the wound (head and face, neck, chest and back, abdomen, upper limb, or lower limb) by clicking or using a drop-down menu, or directly input custom elastic modulus and Poisson's ratio values. The elastic parameter lookup subunit has a built-in anatomical location elastic parameter lookup table described in the method embodiment, automatically querying and returning the corresponding elastic modulus based on the selected anatomical location. Compared to Poisson The radial deformation calculation subunit performs inverse deformation correction on the wound contour according to the radial retraction formula and time decay factor formula described in the method embodiment, and outputs the corrected Fourier descriptor and corrected local geometric parameters. Preferably, the operator can also input the time interval from injury to taking the picture. The system automatically calculates the time decay factor and incorporates it into the deformation compensation calculation.

[0058] The similarity matching module corresponds to step S4 in the method embodiment. It is configured to calculate cosine similarity between the corrected Fourier descriptor and the Fourier descriptors of each tool in the wounding tool cross-section library, fuse the weighted Euclidean distance with corrected local geometric parameters to obtain a comprehensive matching score, and output a recommended list of wounding tool types in descending order. This module includes a cross-section library management subunit, a dual-channel matching calculation subunit, a sorting output subunit, and a feedback correction subunit. The cross-section library management subunit is responsible for the storage, retrieval, and update management of the wounding tool cross-section library, supporting operators to add new tool cross-section samples to expand the coverage of the cross-section library. The dual-channel matching calculation subunit performs global and local dual-channel matching calculations according to the cosine similarity formula and weighted Euclidean distance formula described in the method embodiment. The sorting output subunit sorts the results in descending order of the comprehensive matching score, displaying the top 5 matching results in a list format. Each result includes the tool type name, specifications, comprehensive matching score, and sub-scores for both global and local channels. Preferably, the sorting output subunit also displays a visual comparison of the overlay of the tool cross-section contour ranked first and the corrected wound contour, supplemented by semi-transparent fill and color differentiation, enabling forensic experts to intuitively judge the matching effect. The feedback correction subunit implements the closed-loop feedback correction mechanism described in the method embodiment. When the difference between the matching scores of the first and second ranked items is less than the discrimination threshold, it automatically triggers iterative optimization of the empirical correction coefficient and contact stress estimate in the deformation compensation module.

[0059] The data flow relationship between the above four modules is as follows: the output of the standardized image acquisition module (corrected image and pixel physical resolution) is fed into the contour parameterization module; the output of the contour parameterization module (frequency coefficient set and local geometric parameters) is fed into the deformation compensation module; the output of the deformation compensation module (corrected Fourier descriptor and corrected local geometric parameters) is fed into the similarity matching module; when the similarity matching module triggers feedback correction, it feeds back the candidate tool parameters to the deformation compensation module for iterative optimization. This bidirectional data flow architecture of forward data-driven and backward feedback correction ensures the system's analytical accuracy and robustness when facing complex cases.

[0060] In one embodiment of the present invention, the system is implemented using the Python programming language and runs on a Windows 10 or later operating system environment. The core algorithm relies on open-source computing libraries such as NumPy, SciPy, and OpenCV. The system provides human-computer interaction through a graphical user interface (GUI), with an interface layout comprising four functional areas: an image display area, a parameter setting area, a contour editing area, and a result display area. The image display area supports zooming, panning, and rotating of wound images and can overlay automatically extracted contour curves and corrected contour curves. The parameter setting area provides parameter configuration functions such as anatomical site selection, time interval input, harmonic order setting, and weight coefficient adjustment. The contour editing area provides interactive contour correction tools, including control point dragging, edge segment addition and deletion, and smoothing operations. The result display area displays a sorted list of matching results in tabular form and graphically shows the contour overlay comparison effect.

[0061] Preferably, the system of the present invention also includes a data management module, responsible for the storage, retrieval, and export of case data. After each analysis, the system automatically stores all intermediate data, including the original image, corrected image, contour parameters, and matching results, as well as the final result, in a structured format into a local database, generating a unique case number and timestamp. Operators can retrieve historical analysis records by case number or time range, facilitating comparison and tracing of multiple analysis results for the same case. In addition, the system supports exporting complete analysis reports as PDF files. The report content includes wound images, contour extraction results, comparisons before and after deformation compensation, a matching degree score ranking list, and contour overlay comparison diagrams, which can be directly used as technical support material for forensic identification and attached to the identification report.

[0062] The entire system's end-to-end processing time, from image import to matching result output, does not exceed 10 seconds, meeting the timeliness requirements of forensic on-site investigation and case examination.

[0063] To verify the technical effectiveness of this invention, the following comparative experiment was conducted. The experiment used 30 samples of pigskin lacerations caused by known injury-causing tools as the test dataset, including 10 cases of hammer injuries, 8 cases of wrench injuries, 7 cases of tubular object injuries, and 5 cases of brick corner injuries. The injury-causing tool cross-section library contains 48 tool cross-section samples across 4 major categories. The experiment was run on a laptop computer equipped with an Intel Core i7-12700H processor, 16GB DDR5 memory, and Windows 11 operating system. The software environment consisted of Python 3.10 with NumPy 1.24, SciPy 1.11, and OpenCV 4.8.

[0064] Regarding accuracy, Top-1 accuracy (the proportion of the first-ranked match being the correct injury-causing tool category) and Top-3 accuracy (the proportion of the correct category appearing in the top 3) were used as evaluation metrics. The Top-1 accuracy of the method in this invention was 76.7% (23 / 30), and the Top-3 accuracy was 93.3% (28 / 30). In contrast, the traditional forensic visual comparison method, performed independently by three forensic experts with more than 5 years of experience on the same test set, had an average Top-1 accuracy of 53.3% and a Top-3 accuracy of 73.3%, with a consistency among the three forensic experts (measured by the Fleiss-Kappa coefficient) of only 0.42, which is considered moderate consistency. The Top-1 accuracy of the method in this invention is 23.4 percentage points higher than that of the traditional method, and the Top-3 accuracy is 20.0 percentage points higher. Furthermore, due to the deterministic nature of the algorithm, there is no inter-observer variability problem.

[0065] Regarding the deformation compensation effect, the matching results with and without step S3 deformation compensation were compared. Without deformation compensation, directly matching the Fourier descriptor from step S2 with the cross-section library resulted in a Top-1 accuracy of 60.0% (18 / 30). After introducing deformation compensation, the Top-1 accuracy improved to 76.7%, an increase of 16.7 percentage points. Further analysis revealed that deformation compensation had a particularly significant effect on improving matching in anatomical regions with low elastic modulus (such as the abdomen and neck). In samples from these regions, the Top-1 accuracy increased from 45.5% to 72.7%, an improvement of 27.2 percentage points. This is because tissue retraction is more severe in regions with low elastic modulus, and the deformation compensation model provides a larger correction for these samples, resulting in a more significant compensation effect. In contrast, in regions with high elastic modulus, such as the lower limbs, the retraction is relatively small, and the difference in matching results before and after deformation compensation is relatively small.

[0066] Regarding the effectiveness of feedback correction, the feedback correction mechanism was triggered in 8 out of 30 test samples. Of these, 5 matches that ranked first after feedback correction changed from incorrect to correct, and feedback correction improved edge cases by 62.5%. The average number of iterations for feedback correction was 12.3, with an average additional time of 1.5 seconds, which significantly improved matching accuracy within an acceptable time cost range.

[0067] Regarding the selection of harmonic order, the effects of elliptic Fourier descriptors of orders 10, 15, 20, 25, and 30 on matching accuracy were tested. Experimental results show that when the harmonic order increases from 10 to 20, the Top-1 accuracy steadily increases from 63.3% to 76.7%, while increasing from 20 to 30 only slightly increases the accuracy to 80.0%, but the computation time increases by approximately 40%. Therefore, this invention preferably uses order 20 as the default harmonic order setting, achieving a better balance between matching accuracy and computational efficiency.

[0068] In terms of matching efficiency, the average processing time for a single complete analysis process (steps S1 to S4) is 6.8 seconds, with step S1 taking approximately 1.2 seconds, step S2 approximately 2.1 seconds, step S3 approximately 0.8 seconds, and step S4 approximately 2.7 seconds, which meets the timeliness requirements of forensic case investigation. When the cross-section database size expands from 48 samples to 200 samples, the processing time for step S4 increases to approximately 5.1 seconds, and the total processing time is approximately 9.2 seconds, still within the 10-second timeliness constraint.

[0069] In summary, this invention achieves quantitative comparative analysis between the morphological characteristics of blunt force trauma wounds and the injuring tools through a three-in-one collaborative architecture of wound contour Fourier descriptor parametric characterization, tissue shrinkage deformation compensation correction, and injuring tool cross-section similarity matching. This overcomes the technical defects of low specificity of traditional qualitative descriptions and poor reliability of visual comparison, and provides objective and efficient quantitative reference for forensic inference of injuring objects.

[0070] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A method for forensic bludgeon wound image comparison analysis, characterized in that, Includes the following steps: Step S1, Standardized macro photography and wound image acquisition: Perform standardized macro photography on skin lacerations caused by blunt force trauma from an orthophoto perspective. Place a geometric calibration target around the wound and acquire a high-resolution digital image containing the wound area and the geometric calibration target. Calculate the pixel physical resolution of the image based on the known geometric dimensions of the geometric calibration target, and perform geometric correction on the wound area image using the pixel physical resolution to eliminate lens distortion. Step S2, wound contour extraction and Fourier descriptor parameterization: edge detection is performed on the geometrically corrected wound region image to extract the wound edge contour curve. The wound edge contour curve is parameterized and decomposed using an elliptical Fourier descriptor to obtain a set of frequency coefficients representing the overall shape of the wound. At the same time, the wound angle, wound edge curvature distribution and maximum width-to-length ratio of the wound are extracted as local geometric parameters. Step S3, Tissue retraction deformation compensation and correction: Based on the elastic mechanical properties of skin tissue, a tissue retraction deformation compensation model is constructed. According to the tissue elastic parameters corresponding to the anatomical location of the wound, the frequency coefficient set and local geometric parameters obtained in step S2 are reversely deformed and corrected to generate the corrected Fourier descriptor and corrected local geometric parameters after deformation compensation. Step S4, Similarity Matching and Inference of Wounding Tool Sections: The profiles of each blunt instrument section in the wounding tool section library are parameterized and stored using the same elliptical Fourier descriptor as in Step S2. The cosine similarity is calculated using the corrected Fourier descriptor and the Fourier descriptors of each tool in the section library. The weighted Euclidean distance of the corrected local geometric parameters is then fused to obtain a comprehensive matching score. The recommended list of wounding tool types and their corresponding matching scores are output in descending order of the comprehensive matching score.

2. The method of forensic blunt force trauma wound morphological feature image comparison analysis of claim 1, wherein, In step S1, the geometric calibration target includes at least four non-collinear circular markers with a diameter of 3mm to 5mm; the shooting distance for the standardized macro photography is 15cm to 30cm, the image resolution is not less than 4000×3000 pixels, and the accuracy of the pixel physical resolution is not less than 0.05mm / pixel.

3. The method for comparing and analyzing the morphological features of forensic blunt force trauma wounds according to claim 1, characterized in that, In step S2, the harmonic order of the elliptical Fourier descriptor is 10 to 30; the extraction range of the wound angle is 0° to 180°; and the maximum width-to-length ratio is defined as the ratio of the maximum transverse diameter to the maximum longitudinal diameter of the wound.

4. The method for comparing and analyzing the morphological features of forensic blunt force trauma wounds according to claim 1, characterized in that, In step S3, the tissue elasticity parameters include elastic modulus and Poisson's ratio. The elastic modulus is obtained from a preset anatomical elasticity parameter lookup table based on the anatomical location of the wound. The value of the elastic modulus ranges from 0.1 MPa to 10 MPa.

5. The method for comparing and analyzing the morphological features of forensic blunt force trauma wounds according to claim 1, characterized in that, In step S2, the edge detection adopts multi-scale Canny edge detection, which specifically includes: performing edge detection on the wound area image at three scales with Gaussian kernel standard deviations of 1.0, 2.0 and 4.0 respectively; merging the edge detection results at each scale by performing a logical OR operation; and then using morphological closure operation to eliminate edge breaks and extract the connected wound edge contour curves.

6. The method for comparing and analyzing the morphological features of forensic blunt force trauma wounds according to claim 1, characterized in that, In step S3, the specific method of the inverse deformation correction is as follows: construct a deformation field with radial basis functions, estimate the amount of tissue retraction of the wound contour from the moment of injury to the moment of image acquisition based on the elastic modulus and Poisson's ratio, and perform elliptical Fourier descriptor decomposition again after radially expanding the wound edge contour curve by the amount of tissue retraction.

7. The method for comparing and analyzing the morphological features of forensic blunt force trauma wounds according to claim 1, characterized in that, In step S4, the comprehensive matching score is calculated as follows: the cosine similarity is multiplied by the first weight coefficient to obtain the global shape weighted score; the normalized local geometric parameter weighted Euclidean distance is multiplied by the negative exponent by the second weight coefficient to obtain the local parameter weighted score; the global shape weighted score and the local parameter weighted score are summed, wherein the first weight coefficient is 0.6 to 0.8, and the second weight coefficient is 1 minus the first weight coefficient.

8. The method for comparing and analyzing the morphological features of forensic blunt force trauma wounds according to claim 1, characterized in that, The wounding tool section library includes at least four types of blunt object section profiles: hammer face section, wrench section, tubular object section, and brick corner section. Each type of blunt object section profile includes no fewer than five different sizes of section samples.

9. The method for comparing and analyzing the morphological features of forensic blunt force trauma wounds according to claim 1, characterized in that, Step S4 also includes a feedback correction step: when the difference between the comprehensive matching score of the first-ranked injury tool type and the comprehensive matching score of the second-ranked injury tool type is less than a preset discrimination threshold, the cross-sectional contour parameters of the first-ranked injury tool are input in reverse into the tissue shrinkage deformation compensation model in step S3 for iterative optimization correction, and then the matching calculation in step S4 is re-executed.

10. A forensic blunt force trauma wound morphological feature image comparison and analysis system, used to implement the forensic blunt force trauma wound morphological feature image comparison and analysis method according to any one of claims 1-9, characterized in that, include: A standardized image acquisition module is configured to perform standardized macro photography of skin lacerations caused by blunt force trauma from an orthophoto perspective. A geometric calibration target is placed around the wound, and a high-definition digital image containing the wound area and the geometric calibration target is acquired. The pixel physical resolution is calculated based on the known geometric dimensions of the geometric calibration target to perform geometric correction on the wound area image. The contour parameterization module is configured to perform edge detection on the geometrically corrected wound region image to extract the wound edge contour curve, perform parameterization decomposition of the wound edge contour curve using an elliptical Fourier descriptor to obtain a set of frequency coefficients, and extract the wound angle, wound edge curvature distribution and maximum width-to-length ratio as local geometric parameters. The deformation compensation module is configured to perform inverse deformation correction on the frequency coefficient set and local geometric parameters based on the elastic mechanical properties of skin tissue and the tissue elastic parameters corresponding to the anatomical location of the wound, and generate a corrected Fourier descriptor and corrected local geometric parameters. The similarity matching module is configured to calculate the cosine similarity between the corrected Fourier descriptor and the Fourier descriptor of each tool in the wounding tool section library, fuse the weighted Euclidean distance with corrected local geometric parameters to obtain a comprehensive matching score, and output a recommended list of wounding tool types and their corresponding matching scores in descending order of the comprehensive matching score.

Images