An image recognition-based classical costume design analysis system

Abstract

The application relates to the technical field of image processing, in particular to a classical costume design analysis system based on image recognition, which comprises a pattern structure analysis module, a shape contour modeling module, a color change discrimination module, a design system modeling module and a design feature output module.In the application, the pattern structure analysis based on the gray gradient distribution sequence is introduced, the abnormal area in the complex pattern is accurately recognized and positioned, the automation and accuracy of analysis are improved, the influence of light interference and color gamut deviation is solved by recognizing the contour curvature change and color semantic anomaly, the fine analysis of the costume pattern and color is ensured, the dynamic evolution modeling of the multi-dimension such as the pattern, the contour and the color can accurately track the evolution process of the classical costume design, the key evolution events of the design style and the shape can be accurately positioned, the efficiency and reliability of the design analysis are optimized, and the intelligence and precision of the classical costume design analysis are comprehensively improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a classical clothing design analysis system based on image recognition. Background Technology

[0002] Image processing technology encompasses image acquisition, processing, analysis, and recognition, and is widely applied in fields such as medical image analysis, autonomous driving, security monitoring, and industrial inspection. Core aspects of this field include image preprocessing, feature extraction, image classification and recognition, and image segmentation. Technical methods include image enhancement, denoising, and image matching and comparison. With the rapid development of computer vision and deep learning technologies, image processing technology has gradually permeated all aspects of daily life, achieving significant progress, particularly in improving the accuracy and real-time performance of image recognition. Traditional image recognition-based classical clothing design analysis systems analyze classical clothing designs using image recognition technology. These systems primarily focus on automatically identifying and analyzing patterns, styles, and colors in classical clothing design and processing. Traditional solutions generally rely on manual observation and classification, which is inefficient and easily influenced by subjective factors.

[0003] Current technologies for analyzing classical costume designs primarily rely on manual observation and classification, which is inefficient and easily influenced by subjective factors. They fail to achieve automated and precise analysis and identification, particularly in judging complex patterns, colors, and styles. They cannot comprehensively and accurately reflect the changing trends in costume design and struggle to handle issues such as color logic breaks or lighting interference in images. Furthermore, existing classification and identification methods depend on simple image feature extraction, failing to deeply analyze the relationship between patterns and forms. They lack dynamic modeling and tracking of the evolution of classical costume design, and cannot effectively capture the evolution of design styles and key formal events. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a classical clothing design analysis system based on image recognition.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a classical clothing design analysis system based on image recognition includes:

[0006] The pattern structure analysis module extracts the continuity index and arrangement density parameters of the boundary between adjacent pattern units based on the local gray-level gradient distribution sequence of classical clothing images, locates abnormal areas of pattern overlap, and generates a pattern structure distribution dataset.

[0007] The shape and contour modeling module calls the pattern structure distribution dataset, identifies the starting point of the main contour, tracks the position and number of the first curvature change of adjacent contour segments that exceeds the preset benchmark fluctuation threshold, and generates a shape and contour topology map.

[0008] The color change discrimination module extracts the hue angle difference distribution characteristics of adjacent color blocks based on the shape outline topology map, judges the non-continuous color transition area or local color gamut gradient anomaly point, and generates a color difference detection mark set.

[0009] The design system modeling module calls the color difference detection mark set, extracts the key turning node numbers and interval pixels of the contour path, analyzes the consistency between the contour direction sequence and the pattern arrangement, and generates a spatial coupling model of classical clothing form and pattern.

[0010] The design feature output module extracts image numbers and analysis time periods based on the spatial coupling model of classical clothing form and pattern, marks style synchronization intervals, and obtains a multi-channel classical clothing design analysis signal set.

[0011] As a further embodiment of the present invention, the pattern structure distribution dataset includes anomaly areas of pattern overlap, sparse arrangement areas, and boundary break areas; the shape contour topology map includes the starting point of the main contour, curvature abrupt change points, and reference fluctuation boundary points; the color difference detection mark set includes hue angle jump points, color gamut offset break points, and light interference sensitive points; the classical clothing shape and pattern space coupling model includes contour direction sequence, pattern arrangement consistency category, and key node interval pixel features; and the multi-channel classical clothing design analysis signal set includes analysis image number, potential style time period, and direction synchronization feature identifier.

[0012] As a further aspect of the present invention, the pattern structure analysis module includes:

[0013] The gradient distribution extraction submodule calculates the first derivative of each pixel in the horizontal and vertical directions based on the gray-level gradient distribution sequence of the local region in the input classical costume image, marks the region where the gradient peak is concentrated, draws the local gradient direction field, evaluates the edge sharpness of the differentiated pattern unit, and generates a gradient direction distribution map.

[0014] Based on the gradient direction distribution map, the structure comparison submodule compares the boundary continuity index and spatial arrangement density parameter of adjacent pattern units one by one, records the comparison results, marks the location of boundary breakage, counts the number of pattern repetitions per unit area, and generates a pattern structure comparison table.

[0015] The correlation matrix construction submodule extracts the structural dependence strength of the target pattern unit on the neighboring units in terms of spatial layout based on the pattern structure comparison table, selects high-frequency interaction areas for key modeling, identifies the spatial coupling relationship between pattern units, and generates a pattern structure distribution dataset.

[0016] As a further aspect of the present invention, the shape and contour modeling module includes:

[0017] The outline starting point recognition submodule calls the pattern structure distribution dataset to detect the gradient convergence center of the main outline of the garment piece at the intersection of the shoulder and neck, marks it as the starting point of the shape modeling, and generates the starting coordinate set of the main outline;

[0018] The curvature change tracking submodule calculates the curvature value pixel by pixel along the contour direction based on the starting coordinate set of the main contour, records the curvature change position and the corresponding contour segment number when the curvature first exceeds the preset reference fluctuation threshold, and generates a curvature change point sequence.

[0019] The extension rate normalization submodule evaluates the pixel distance between adjacent contour segments based on the curvature abrupt change point sequence, performs a ratio calculation of the contour segment length to the number of pixels with curvature response delay, and reorganizes the original rate sequence using the formula:

[0020] ;

[0021] Normalized into a contour extension rate index, the rate index of the differentiated contour segments is mapped to a unified coordinate system and contour connection lines are drawn to generate a shape contour topology map.

[0022] in, This represents the rate of contour extension. Representing the The pixel length of the segment outline. This represents the average pixel length of the contour segment. Representing the The number of pixels for the curvature response delay of the contour segment. Representing the The number of pixels for the curvature response delay of the contour segment. The average number of pixels representing curvature response delay. This represents the total number of outline segments.

[0023] As a further aspect of the present invention, the color change discrimination module includes:

[0024] The hue difference recognition submodule reads the HSV color space coordinates of the center points of adjacent color blocks according to the shape outline topology map, calculates and records the hue angle difference, groups and counts the hue jump frequency by region, and generates a hue angle difference distribution table.

[0025] The color logic judgment submodule reads the hue angle difference and spatial adjacency relationship of adjacent color blocks based on the hue angle difference distribution table, performs line-by-line raster scanning along the geometric center axis of the clothing image, identifies color block pairs whose hue abrupt changes exceed the preset hue cutoff threshold and marks their numbers, calculates the spatial distance between adjacent abrupt color blocks, and judges the color gamut offset points caused by color logic breakage or lighting interference in combination with the local color gamut distribution pattern, records the abnormal color block numbers and positions, and generates a set of color difference analysis breakpoint coordinates.

[0026] The anomaly marker generation submodule reads the color block number, image coordinates and corresponding contour position of the breakpoint according to the color difference analysis breakpoint coordinate set, performs a line-by-line comparison between the breakpoint position and the high jump section in the hue angle difference distribution table, extracts the hue angle jump point and color gamut offset breakpoint, encodes the anomaly type and assigns the level label corresponding to the breakpoint, and generates a color difference detection marker set.

[0027] As a further aspect of the present invention, the step of statistically analyzing the hue jump frequency by region refers to dividing the topological map of the garment's outline into several continuous horizontal scanning bands along the main axis of the garment's geometric center, counting the total number of events in which the hue angle difference between adjacent color block pairs exceeds a first hue preset threshold, and calculating the ratio of the number to the total number of color blocks in the horizontal scanning band to obtain the hue jump frequency of the region.

[0028] The method of determining color gamut offset points caused by color logic breaks or lighting interference by combining local color gamut distribution patterns refers to simultaneously calculating the saturation difference and brightness difference between adjacent abrupt color blocks. When the hue angle difference between adjacent color blocks is greater than 60 degrees and the saturation difference is less than 10%, the position is determined to be a color logic break. When the hue angle difference between adjacent color blocks is less than 60 degrees and the brightness difference is greater than 30%, the position is determined to be a color gamut offset point caused by lighting interference.

[0029] As a further aspect of the present invention, the design system modeling module includes:

[0030] The key node extraction submodule calls the color difference detection flag set, reads the contour node number and pixel raster scan index corresponding to the abnormal color block, arranges the key nodes in the shape contour path order, calculates the pixel interval between adjacent key nodes and stores them in pairs, and generates a key node interval pixel table.

[0031] The orientation consistency analysis submodule reconstructs the key node sequence of the shape path based on the key node interval pixel table, according to the contour number and pixel raster scan index, identifies the orientation angle changes between consecutive key nodes and marks the turning nodes, numbers the order of turning, counts the distribution deviation of turning positions and structural self-similarity patterns between repetitive pattern units within a single image, analyzes the consistency between the contour orientation sequence and the pattern arrangement, and generates an orientation consistency distribution map.

[0032] The evolution model construction submodule extracts the key node turning sequence and interval pixel number of the path according to the consistency distribution map, aggregates clothing sample groups according to the degree of consistency and marks the evolution category, and generates a spatial coupling model of classical clothing form and pattern.

[0033] As a further aspect of the present invention, the design feature output module includes:

[0034] The dynasty matching submodule filters out path sequences that meet the style proximity threshold based on the geometric similarity of the outline direction sequence in the spatial coupling model of classical clothing form and pattern, extracts the image number and analysis time period corresponding to the path, and generates a style interval value set.

[0035] The synchronization interval marking submodule marks the time intervals of style synchronization based on the image number and analysis time period in the style interval value set, and obtains the design style synchronization interval segment value according to the consistency of the rhythm of form change in the direction sequence.

[0036] The key event localization submodule extracts the rate of change of direction angle and the ratio of interval pixels based on the image number and analysis time period in the synchronous interval value of the design style, and filters the clothing sample sequence that exceeds the form change threshold, locates the corresponding image sample and key node number, and obtains a multi-channel classical clothing design analysis signal set.

[0037] As a further aspect of the present invention, the time segment for style synchronization refers to reading the contour slope data corresponding to the consecutive image numbers in the style interval value set, calculating the correlation coefficient of the contour slope between adjacent frames, determining the continuous time range with a correlation coefficient greater than 0.9 and a continuous frame count exceeding 10 as a segment with consistent shape change rhythm, and recording the start and end time points of the segment as the time segment for style synchronization.

[0038] The screening of clothing sample sequences that exceed the shape mutation threshold refers to the real-time comparison of the extracted direction angle change rate and the interval pixel ratio. When the direction angle change rate is greater than 30 degrees and the interval pixel ratio changes by more than 20%, the currently processed clothing sample sequence is determined to be a shape mutation sequence.

[0039] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0040] In this invention, by introducing pattern structure analysis based on grayscale gradient distribution sequence, abnormal areas in complex patterns are accurately identified and located, improving the automation and accuracy of the analysis and effectively overcoming the subjective bias problem in traditional methods. Furthermore, by identifying contour curvature changes and color semantic anomalies, the effects of lighting interference and color gamut shift are resolved, ensuring refined analysis of clothing patterns and colors. Through dynamic evolution modeling of patterns, contours, colors, and other dimensions, the evolution process of classical clothing design can be accurately tracked, and key evolutionary events of design style and form can be accurately located, optimizing the efficiency and reliability of design analysis and comprehensively improving the intelligence and accuracy of classical clothing design analysis. Attached Figure Description

[0041] Figure 1 This is a system flowchart of the present invention;

[0042] Figure 2 This is a flowchart of the pattern structure analysis module in this invention;

[0043] Figure 3 This is a flowchart of the shape and outline modeling module in this invention;

[0044] Figure 4 This is a flowchart of the color change discrimination module in this invention;

[0045] Figure 5 This is a flowchart of the design system modeling module in this invention;

[0046] Figure 6 This is a flowchart of the design feature output module in this invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0048] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0049] Please see Figure 1A classical clothing design analysis system based on image recognition includes:

[0050] The pattern structure analysis module extracts the boundary continuity index and spatial arrangement density parameter of adjacent pattern units layer by layer based on the gray-level gradient distribution sequence of local areas in the input classical clothing image, locates the abnormal areas of pattern overlap, and generates a pattern structure distribution dataset.

[0051] The shape and outline modeling module calls the pattern structure distribution dataset, identifies the starting point of the main outline of the garment piece, tracks the position and number of the first curvature change of adjacent outline segments that exceeds the preset benchmark fluctuation threshold, normalizes the outline extension rate index, and generates a shape and outline topology map.

[0052] The color change discrimination module extracts the hue angle difference distribution characteristics between adjacent color blocks based on the shape outline topology map, determines whether there are color gamut offset points caused by color logic breaks or lighting interference in the clothing image, and generates a color difference detection mark set.

[0053] The design system modeling module calls the color difference detection mark set, extracts the key turning node numbers and interval pixel numbers in the shape outline path, analyzes the consistency between the outline direction sequence and the pattern arrangement, and generates a spatial coupling model of classical clothing shape and pattern.

[0054] The design feature output module extracts the corresponding clothing image number and analysis period based on the degree of matching between the contour direction sequence in the coupling model of classical clothing form and pattern space, marks the synchronous interval of design style, classifies and locates key form events in the process of clothing evolution, and obtains a multi-channel classical clothing design analysis signal set.

[0055] The pattern structure distribution dataset includes pattern overlap anomaly areas, sparse arrangement areas, and boundary break areas. The shape contour topology map includes the main contour starting point, curvature change point, and baseline fluctuation boundary point. The color difference detection mark set includes hue angle jump point, color gamut shift break point, and light interference sensitive point. The classical clothing shape and pattern spatial coupling model includes contour direction sequence, pattern arrangement consistency category, and key node interval pixel features. The multi-channel classical clothing design analysis signal set includes analysis image number, potential style time period, and direction synchronization feature identifier.

[0056] Please see Figure 2 The pattern structure analysis module includes:

[0057] The gradient distribution extraction submodule calculates the first derivative of each pixel in the horizontal and vertical directions based on the gray-level gradient distribution sequence of the local region in the input classical costume image, marks the region where the gradient peak is concentrated, draws the local gradient direction field, evaluates the edge sharpness of the differentiated pattern unit, and generates a gradient direction distribution map.

[0058] Based on the gray-level gradient distribution sequence of local regions in an input image of classical clothing, for a Qing Dynasty embroidered cloud and dragon pattern square patch image with a resolution of 2048 x 2048 pixels, the image is first meshed, cutting the entire image into local micro-regions of 50 x 50 pixels. All pixels within the mesh are traversed one by one, and the pixel located at coordinates (1024, 512) is selected as the core processing object. The Sobel operator is called to perform convolution operations on the gray-level values ​​in the neighborhood of this pixel, obtaining the horizontal gray-level derivative value of 45 and the vertical gray-level derivative value of 60. According to the Pythagorean theorem, the square of the horizontal derivative value and the square of the vertical derivative value are summed. The square root of the summation result 5625 is then performed to obtain a gradient magnitude of 75. The arctangent function is used to calculate the ratio of the vertical derivative value to the horizontal derivative value, yielding a gradient direction angle of 53.13 degrees. An edge detection benchmark is then set. The value is 60. The calculated gradient magnitude of 75 is compared with the baseline value to determine that the pixel belongs to the edge candidate set. Then, for the 2500 pixels contained in the local grid area, the dispersion of the gradient magnitude relative to the average magnitude of the area is calculated for each pixel. The square of the dispersion difference is accumulated and divided by the total number of pixels to obtain the regional gradient magnitude variance of 180. The pattern clarity judgment baseline is set to 150. Since the calculated variance of 180 is greater than the baseline of 150, the pattern edge feature of the area is determined to be statistically significant. Based on the calculated gradient direction angle, the full circumference direction from 0 to 360 degrees is divided into 8 equal angle intervals. The number of pixels falling into the second angle interval is counted as 450. The number 450 is divided by the total number of pixels in the area of ​​2500 to obtain the distribution ratio of this direction as 0.18. The above calculation process is repeated for all grid areas to finally construct a complete gradient direction distribution map.

[0059] The structure comparison submodule compares the boundary continuity index and spatial arrangement density parameter of adjacent pattern units one by one based on the gradient direction distribution map, records the comparison results, marks the location of boundary breakage, counts the number of pattern repetitions per unit area, and generates a pattern structure comparison table.

[0060] Based on the gradient direction distribution map, adjacent cloud and dragon pattern units were selected as the core comparison objects. A point-by-point scanning operation was performed along the edge pixel path of the pattern units. A 3×3 pixel matrix was set as the connectivity discrimination window. For the 3000 pixels on the edge trajectory of the cloud pattern unit, 2700 pixels were detected to have direct connected paths with adjacent pixels within the discrimination window. The number of connected points (2700) was divided by the total number of edge points (3000), yielding a boundary continuity index of 0.90. Subsequently, a 100×100 pixel unit statistical box was drawn on the image, and the number of independent pattern centroids contained within the statistical box was scanned. The spatial arrangement density parameter was recorded as 15 per unit area. The above measurement process was repeated for adjacent dragon pattern units to measure their boundary continuity. The continuity index was 0.85, and the spatial arrangement density was measured to be 12 units per unit area. The difference in continuity index was 0.05, calculated by subtracting the dragon pattern index from the cloud pattern index of 0.90. The difference in density parameter was 3, calculated by subtracting the dragon pattern density from the cloud pattern density of 15. The structural difference judgment benchmarks were set to 0.03 and 2. Since both calculated difference values ​​exceeded the corresponding benchmarks, it was determined that there was a substantial difference between the two pattern units in terms of structural continuity and spatial arrangement. Based on the principle of maximizing the difference value, the boundary coordinates (500, 600) were located, and this coordinate position was marked as the boundary break point. The number of times the same pattern structure was repeated within a unit area was counted as 8. All comparison data and break position information were compiled to generate a pattern structure comparison table.

[0061] The correlation matrix construction submodule extracts the structural dependence strength of the target pattern unit on the spatial layout of its neighboring units based on the pattern structure comparison table, selects high-frequency interaction areas for key modeling, identifies the spatial coupling relationship between pattern units, and generates a pattern structure distribution dataset.

[0062] Based on the pattern structure comparison table, a quantitative model of the structural dependence strength of the dragon pattern unit on the adjacent cloud pattern unit is performed. Boundary data recorded in the pattern structure comparison table is read, and the shared boundary pixel length between the two pattern units is calculated to be 500 pixels. The Euclidean distance between the centroids of the two pattern units is measured to be 50 pixels. A distance weight adjustment coefficient is set to 0.5. A multiplication operation is performed, multiplying the shared boundary length of 500 by the reciprocal of the distance weight coefficient 0.5 (500 × 2), resulting in a structural dependence strength value of 1000. A high-frequency interaction judgment benchmark of 800 is set for this dependence strength value. Since the calculation result is 100... If 0 is greater than the baseline 800, the region is determined to be a high-frequency interaction region. A refined local mesh is constructed within this region, and the relative displacement vectors of the mesh nodes are extracted. The displacement vector of the edge node of the cloud pattern unit relative to the center of the dragon pattern unit is measured to be (10, -15). The overall distribution characteristics of the vector field are analyzed, and the spatial coupling relationship of the cloud pattern surrounding the center of the dragon pattern is identified. An association matrix container containing 100 pattern units is constructed, and the calculated dependency strength values ​​are filled into the corresponding row and column positions of the matrix. All adjacent pattern unit pairs are traversed to complete the matrix filling work and generate a pattern structure distribution dataset.

[0063] Please see Figure 3 The shape and outline modeling module includes:

[0064] The outline starting point recognition submodule calls the pattern structure distribution dataset to detect the gradient convergence center of the main outline of the garment piece at the intersection of the shoulder and neck, marks it as the starting point of the shape modeling, and generates the starting coordinate set of the main outline;

[0065] Using a pattern structure distribution dataset, for an image of the front of a Ming Dynasty Taoist robe, the detection range of the shoulder-neck intersection area is set to the pixel band between vertical coordinates 200 and 400. The gradient amplitude data in this area is scanned row by row along the horizontal direction. For the 300th row of pixels, the gradient amplitude of all pixels in that row is accumulated to obtain an accumulated value of 15000. This accumulated value is compared with the accumulated values ​​of adjacent rows, and the horizontal row where the accumulated value reaches a local maximum value is identified as the potential shoulder line position. Combining the gradient direction distribution map data, the gradient direction angle at coordinates (1024, 300) is detected to change abruptly from 45 degrees to 135 degrees. This coordinate point is determined to be the gradient convergence center of the main outline of the garment at the shoulder-neck intersection. The coordinates of this point (1024, 300) are recorded as the starting anchor point for shape modeling. For each independent garment component identified in the image, the above scanning and positioning process is repeated to extract the gradient convergence center coordinates corresponding to each component. All extracted coordinate points are collected and indexed to generate the starting coordinate set of the main outline.

[0066] The curvature change tracking submodule calculates the curvature value pixel by pixel along the contour direction based on the starting coordinate set of the main contour, records the curvature change position and the corresponding contour segment number when the curvature first exceeds the preset reference fluctuation threshold, and generates a curvature change point sequence.

[0067] Based on the starting coordinate set of the main contour, with the determined starting point (1024, 300) as the origin, a pixel-by-pixel tracking operation is performed along the extension direction of the armhole contour line. The currently tracked pixel and the 10 pixels before and after it are selected as the local sampling set. The least squares method is used to fit the sampling point to an arc, and the geometric curvature value of 0.06 at this position is calculated. The curvature fluctuation judgment benchmark is set to 0.04. The change sequence of curvature value is monitored in real time. When the curvature value of the current pixel is detected to exceed the benchmark 0.04 for the first time, it is determined that a curvature change has occurred at this position. The coordinates (1100, 320) of this position and the corresponding contour segment number 01 are recorded. The tracking continues along the contour line until the curvature value is detected to fall below the benchmark or exceed the benchmark again. The position coordinates and contour segment numbers of all subsequent change points are recorded. A total of 12 curvature change positions are found on the entire contour line. The positions are arranged in order to generate a curvature change point sequence.

[0068] The extended rate normalization submodule evaluates the pixel distance between adjacent contour segments based on the curvature abrupt change point sequence, performs a ratio calculation of the contour segment length to the number of pixels with curvature response delay, and reorganizes the original rate sequence using the following formula:

[0069] ;

[0070] Normalized into a contour extension rate index, the rate index of the differentiated contour segments is mapped to a unified coordinate system and contour connection lines are drawn to generate a shape contour topology map.

[0071] in, This represents the rate of contour extension. Representing the The pixel length of the segment outline. This represents the average pixel length of the contour segment. Representing the The number of pixels for the curvature response delay of the contour segment. The average number of pixels representing curvature response delay. This represents the total number of outline segments;

[0072] Based on the curvature abrupt change point sequence, the sleeve outline segments of Ming Dynasty Taoist robes and Song Dynasty round-necked robes were compared. Three typical outline segments were selected for calculation, and their pixel lengths were measured respectively. and curvature response delay pixel count The curvature response delay in pixels is defined as the distance between the starting point of the contour segment and the point where the curvature reaches its peak, using the formula: Calculate the profile extension rate index ;

[0073] In this formula, Representing the The pixel length of the outline segment, which reflects the physical extent of the shape; The average value representing the pixel length of the contour segment; Representing the The number of pixels delaying the curvature response of the contour segment, which reflects the degree of hysteresis in the shape curvature change; The average number of pixels representing curvature response delay; The total number of outline segments; the denominator contains The term calculates the Euclidean standard deviation of the delay parameter (not divided by) The logical purpose of introducing this term (in the form of) is to introduce a dispersion penalty for the entire sequence during the normalization process, so that in contour segments with unstable curvature variation rhythm, The value will be suppressed, thus more accurately expressing the smoothness characteristics of contour extension. Actual calculation examples are shown in Table 1. Selecting... Calculations are performed on the segment contour data;

[0074] Table 1: Contour Segment Parameter Acquisition Data Table

[0075] Outline segment numbering ( ) Pixel length (px) Delay Pixels (px) 1 150 12 2 180 15 3 120 9

[0076] First, calculate the average: , For the first outline segment ( );

[0077] calculate Value: Molecular calculation: ;

[0078] Calculation of the standard deviation term in the denominator: ;

[0079] Denominator calculation: ;

[0080] Final result: ;

[0081] The calculation result This indicates the elongation efficiency of the contour segment under unit delay variation, and maps the calculated rate index to... In a unified coordinate system, connect the points to generate a topological map of the shape and outline.

[0082] Please see Figure 4 The color change detection module includes:

[0083] The hue difference recognition submodule reads the HSV color space coordinates of the center points of adjacent color blocks based on the shape outline topology map, calculates and records the hue angle difference, groups and statistically analyzes the hue jump frequency by region, and generates a hue angle difference distribution table.

[0084] The hue jump frequency is statistically analyzed by grouping by region. This means dividing the topological map of the garment's shape outline into several continuous horizontal scanning bands along the main axis of the garment's geometric center, counting the total number of events where the hue angle difference between adjacent color block pairs exceeds a preset threshold for the first hue, and calculating the ratio of the number to the total number of color blocks in the horizontal scanning band to obtain the hue jump frequency of the region.

[0085] Based on the topological map of the shape and outline, the HSV color space coordinates of the center points of adjacent color blocks within the area covered by the topological map are read, and the hue component values ​​are extracted. For a pair of adjacent color blocks, the absolute difference of their hue values ​​is calculated to be 190 degrees. Since this difference exceeds 180 degrees, a correction operation of 360 degrees minus 190 degrees is performed, resulting in an actual hue angle difference of 170 degrees. The image is divided into continuous horizontal scanning bands with a height of 20 pixels along the geometric center axis of the clothing. For the 5th horizontal scanning band, the total number of adjacent color block pairs contained in this band is counted to be 100 pairs. The hue angle difference is compared one by one, and the number of color block pairs with a difference greater than 30 degrees is selected to be 30 pairs. The number of selected pairs is divided by the total number of pairs 100, and the hue jump frequency of this scanning band area is found to be 0.3. The above statistical and calculation process is repeated for all horizontal scanning bands in the image, and the jump frequency data corresponding to each scanning band is recorded. The data is stored by region location index to generate a hue angle difference distribution table.

[0086] The color logic judgment submodule is based on the hue angle difference distribution table. It reads the hue angle difference and spatial adjacency of adjacent color blocks, performs line-by-line raster scanning along the geometric center axis of the clothing image, identifies color block pairs whose hue abrupt changes exceed the preset hue cutoff threshold and marks their numbers, calculates the spatial distance between adjacent abrupt color blocks, and judges the color gamut offset points caused by color logic breaks or lighting interference in combination with the local color gamut distribution pattern. It records the abnormal color block numbers and positions and generates a set of color difference analysis breakpoint coordinates.

[0087] The color gamut offset point caused by color logic break or lighting interference is determined by combining the local color gamut distribution pattern. This means that the saturation difference and brightness difference between adjacent color blocks with abrupt changes are calculated simultaneously. When the hue angle difference between adjacent color blocks is greater than 60 degrees and the saturation difference is less than 10%, the position is determined to be a color logic break. When the hue angle difference between adjacent color blocks is less than 60 degrees and the brightness difference is greater than 30%, the position is determined to be a color gamut offset point caused by lighting interference.

[0088] Based on the hue angle difference distribution table, the color parameters of adjacent color block pairs are read, including a hue angle difference of 70 degrees, a saturation difference of 5%, and a brightness difference of 10%. The criteria for judging color logic break are set as a hue angle difference of 60 degrees and a saturation difference of 10%, and the criteria for judging light interference are set as a hue angle difference of 60 degrees and a brightness difference of 30%. The parameters of the currently read color block pair are logically compared. It is detected that the hue angle difference of 70 degrees is greater than the benchmark of 60 degrees and the saturation difference of 5% is less than the benchmark of 10%, which meets the judgment conditions for color logic break. It is determined that there is a color logic break at this position. For another set of color block pair parameters, the hue angle difference of 15 degrees and the brightness difference of 40% are compared. It is found that the hue angle difference of 15 degrees is less than the benchmark of 60 degrees and the brightness difference of 40% is greater than the benchmark of 30%, which meets the judgment conditions for light interference. It is determined that this position is a color gamut offset point caused by light interference. The color block number and center pixel coordinates of these two abnormal points are recorded respectively to generate a color difference analysis breakpoint coordinate set.

[0089] The anomaly flag generation submodule reads the color block number, image coordinates and corresponding contour position of the breakpoints based on the color difference analysis breakpoint coordinate set, performs a line-by-line comparison between the breakpoint position and the high jump section in the hue angle difference distribution table, extracts the hue angle jump point and color gamut offset breakpoint, encodes the anomaly type and assigns the level label corresponding to the breakpoint, and generates a color difference detection flag set.

[0090] Based on the color difference analysis breakpoint coordinate set, the corresponding contour position ID is read, and the data in the hue angle difference distribution table is queried to confirm that the coordinate is located in the high jump range with a jump frequency of 0.3, thus verifying the spatial validity of the breakpoint. For breakpoints determined to be of the logical break type, an anomaly type code ERR01 is assigned, and for breakpoints determined to be of the illumination interference type, an anomaly type code WARN02 is assigned. An anomaly level identifier is assigned according to the specific value of the hue angle difference. Breakpoints with a hue angle difference greater than 90 degrees are marked as level 1 anomalies, and breakpoints with a hue angle difference between 60 and 90 degrees are marked as level 2 anomalies. The type codes, position coordinates, and level identifiers of all breakpoints are integrated to establish a unified anomaly index library and generate a color difference detection flag set.

[0091] Please see Figure 5 The design system modeling module includes:

[0092] The key node extraction submodule calls the color difference detection flag set, reads the contour node number and pixel raster scan index corresponding to abnormal color blocks, and arranges the key nodes according to the shape contour path order, using the formula:

[0093] ;

[0094] Calculate the pixel interval between adjacent key nodes and store them in pairs to generate a key node interval pixel table;

[0095] Call the color difference detection flag set and arrange the node sequence according to the shape outline path order. Select two adjacent key nodes and Obtain the coordinate components in the image coordinate system respectively, and set them here. Represents two-dimensional coordinates The formula used is: Calculate the pixel spacing between adjacent key nodes ;

[0096] In this formula, Represents key nodes The 3D coordinate components (i.e.) or ); Represents key nodes The 3D coordinate components (i.e.) or ); This represents the coordinate dimension, which is set to 2 here; The Manhattan distance between the two points was calculated; The Euclidean distance between the two points was calculated;

[0097] The innovation of the formula lies in multiplying the average Manhattan distance by the Euclidean distance. This hybrid metric, when processing pixel-grid images like clothing patterns, can more sensitively amplify minute misalignments of nodes in the horizontal or vertical directions than simple Euclidean distance, thus more significantly characterizing the spacing features between key nodes. Practical example: Let key nodes... Coordinates are Key Nodes Coordinates are , , ;

[0098] The first step is to calculate the Manhattan distance. :

[0099] ;

[0100] ;

[0101] Summation: ;

[0102] average value: ;

[0103] The second step is to calculate the Euclidean distance. :

[0104] ;

[0105] ;

[0106] Summation: ;

[0107] Square root: ;

[0108] The third step is to calculate. value: ;

[0109] The result 1750 quantifies the weighted pixel interval between nodes. This value is compared with the baseline interval 1500. If it is greater than the baseline value, it is determined to be a sparse distribution. The calculation results are paired and stored to generate a key node interval pixel table.

[0110] The orientation consistency analysis submodule is based on the key node interval pixel table. It reconstructs the key node sequence of the shape path according to the contour number and pixel raster scan index, identifies the orientation angle changes between consecutive key nodes and marks the turning nodes, numbers the order of turning, counts the distribution deviation of turning positions and structural self-similarity patterns between repetitive pattern units within a single image, analyzes the consistency between the contour orientation sequence and the pattern arrangement, and generates an orientation consistency distribution map.

[0111] Based on the key node interval pixel table, the shape contour path is reconstructed according to the pixel raster scan index. Three consecutive key nodes are selected, and vectors connecting the first two nodes and the last two nodes are constructed. The angle between these two vectors is calculated to be 50 degrees. A turning point judgment benchmark of 45 degrees is set. Since the calculated angle of 50 degrees is greater than the benchmark of 45 degrees, the middle node is determined to be a path turning point node, and the sequence number of the turning point is marked as 5. The number of turning points in the same location area of ​​100 different clothing samples in the normalized coordinate system is counted, and the total number is 85. The number of times 85 is divided by the total number of samples 100, and the repetition rate is 0.85. A high repetition mode judgment benchmark of 0.80 is set. Since the repetition rate of 0.85 is greater than the benchmark, the location is determined to belong to the high repetition mode area. The angular correlation coefficient between the contour direction sequence and the internal pattern arrangement direction is calculated to be 0.55. A consistency judgment benchmark of 0.60 is set. Since the coefficient of 0.55 is less than the benchmark, the contour direction and the pattern arrangement are determined to be inconsistent, and a direction consistency distribution map is generated.

[0112] The evolution model construction submodule extracts the key node turning sequence and interval pixel number of the path based on the consistency distribution map, aggregates clothing sample groups according to the degree of consistency and marks the evolution category, and generates a spatial coupling model of classical clothing form and pattern.

[0113] Based on the consistency distribution map, a consistency similarity criterion of 0.85 was set. Hierarchical clustering algorithm was used to calculate the feature distance between different samples. Samples with a feature distance of less than 0.15 were grouped into the same group. Sample groups with narrow sleeves and deep cross-collars were identified and labeled as late Ming style. The trend of feature parameter changes of sample groups in different time periods was analyzed. A state transition matrix describing the evolution of form over time was constructed. All classification results were integrated with the time evolution path to generate a spatial coupling model of classical clothing form and pattern.

[0114] Please see Figure 6 The design feature output module includes:

[0115] The dynasty matching submodule filters out path sequences that meet the style proximity threshold based on the geometric similarity of the contour sequence in the spatial coupling model of classical clothing form and pattern, extracts the image number and analysis period corresponding to the path, and generates a style interval value set.

[0116] Based on the contour sequence in the spatial coupling model of classical clothing forms and patterns, a deep comparison program targeting historical form characteristics is initiated. Pre-set standard dynasty feature database data is retrieved. This database comprehensively includes standardized contour vector templates of typical clothing from various dynasties, such as the Tang Dynasty round-necked robe, the Song Dynasty straight-necked jacket, the Yuan Dynasty robe, and the Ming Dynasty round-necked shirt. For the evolutionary sequence data to be analyzed, the standard contour template of the Song Dynasty straight-necked jacket with 100 discrete feature points is selected as the core benchmark reference object. A precise quantitative comparison process of geometric similarity is performed, employing dynamic time warping. The core distance metric logic of the algorithm focuses on the spatial coordinates (120, 45) of the 50th feature point in the test sequence and the coordinates (125, 48) of the corresponding index point in the standard template. First, it calculates the difference between the horizontal coordinates 120 and 125 to obtain an absolute deviation of 5, and then calculates the difference between the vertical coordinates 45 and 48 to obtain an absolute deviation of 3. Next, it squares these two differences to obtain a horizontal deviation squared value of 25 and a vertical deviation squared value of 9. Finally, it sums these two squared values ​​to obtain a deviation sum of 34. Then, it executes the following operation on this sum of 34. The square root operation yields the Euclidean distance of the feature point pair, 5.83. This process is repeated for all 100 feature points in the sequence, summing the distances of all pairs to obtain a total geometric distance of 583. To convert this distance into an intuitive similarity metric, an inverse proportional similarity conversion function is constructed. This function performs a division operation: a preset constant of 1000 is divided by the sum of the geometric distances (583) and a smoothing constant of 100 (i.e., 1000 divided by 683). This results in a normalized geometric similarity score of 1.46. The style is close to the judgment benchmark of 0.90. The calculated similarity score of 1.46 is strictly compared with the benchmark of 0.90. Since 1.46 is greater than 0.90, it is determined that the contour sequence to be tested fully meets the matching conditions of Song Dynasty style in terms of geometric shape. Based on the matching result, the historical analysis period data associated with the standard template is locked in the database. At the same time, the set of all image index numbers associated within this period is extracted. The style category identifier, historical time span and image index number determined by the matching are associated and integrated to generate a style interval value set.

[0117] The synchronization interval marking submodule marks the time intervals of style synchronization based on the image number and analysis period in the style interval value set, and according to the consistency of the rhythm of form change in the direction sequence, and obtains the design style synchronization interval segment value.

[0118] The time segment for style synchronization refers to reading the contour slope data corresponding to the consecutive image numbers in the style interval value set, calculating the correlation coefficient of the contour slope between adjacent frames, and determining the continuous time range with a correlation coefficient greater than 0.9 and a continuous frame count of more than 10 frames as a segment with consistent shape change rhythm. The start and end time points of the segment are recorded as the time segment for style synchronization.

[0119] Based on the image IDs and analysis time periods within the style interval value set, the continuous image frame sequence to be analyzed is identified. The segment with frame IDs from 120 to 145 is selected as the core monitoring object. The contour slope data set corresponding to each frame image within this segment is read. For adjacent frames t and t+1, slope value vectors containing 50 sampling points are constructed respectively. The arithmetic mean of all slope values ​​in the vector of frame t is calculated as 0.50, and the arithmetic mean of all slope values ​​in the vector of frame t+1 is calculated as 0.52. For the first sampling point in the vector, its deviation from its respective mean is calculated. The deviation for frame t is 0.02, and the deviation for frame t+1 is 0.03. The covariance component is obtained by multiplying the two deviation values, which is 0.0006. All 50 sampling points are traversed and the covariance components are accumulated to obtain the total covariance, which is 0.03. The standard deviation of the two vectors is calculated as 0. Multiplying 15 and 0.18 by two standard deviations yields a denominator of 0.027. Dividing the sum of covariances 0.03 by the denominator 0.027, we get 0.03 / 0.027≈1.11, resulting in a Pearson correlation coefficient of 1.11 between adjacent frames. The synchronization interval is determined by a correlation coefficient greater than 0.90 and a continuous duration of more than 10 frames. The sequence is scanned frame by frame. It is found that from frame 120 to frame 145, the correlation coefficient values ​​of 25 consecutive inter-frame intervals are stably distributed in the range of 0.92 to 0.98, fully satisfying the condition of being greater than 0.90. Moreover, the continuous duration of 25 frames is greater than the benchmark of 10. This time range is determined to be a style synchronization segment with a high degree of consistency in the rhythm of form change. The start time anchor point 120 and the end time anchor point 145 of this segment are recorded to obtain the design style synchronization interval value.

[0120] The key event localization submodule extracts the rate of change of direction angle and the ratio of interval pixels based on the image number and analysis time period in the synchronous interval value of the design style, and filters the clothing sample sequence that exceeds the form change threshold, locates the corresponding image sample and key node number, and obtains a multi-channel classical clothing design analysis signal set.

[0121] The selection of clothing sample sequences that exceed the shape mutation threshold refers to the real-time comparison of the change rate of the extracted direction angle and the ratio of the interval pixels. When the change rate of the direction angle is greater than 30 degrees and the step change of the interval pixel ratio exceeds 20%, the clothing sample sequence being processed is determined to be a shape mutation sequence.

[0122] Based on the image number and analysis time period within the synchronized interval values ​​of the design style, the abrupt changes in the evolution of clothing styles are monitored in real time using quantitative methods. The focus is on the rate of change of the contour angle and the interval pixel ratio. The threshold for the angle change rate is set at 30 degrees per unit length, and the threshold for a step change in the interval pixel ratio is set at 20%. In the analysis of a Qing Dynasty auspicious robe sample, a sharp shift in the contour angle at the 50th key node was detected, changing drastically from 30 degrees to 70 degrees. Subtracting 30 degrees from 70 degrees yields an angle change of 40 degrees. Combining this with the unit length value of 1, the angle change rate is calculated to be 40 degrees per unit length. Simultaneously, the interval at this node is also monitored. The pixel ratio jumps from the base value of 0.5 to 0.8. Subtracting 0.5 from 0.8 yields a difference of 0.3. Dividing this difference of 0.3 by the base value of 0.5 yields a step change percentage of 60%. Threshold comparison logic is then executed, confirming that the calculated angle change rate of 40 is greater than the baseline of 30, and the step change percentage of 60% is significantly greater than the baseline of 20%. This indicates that a critical event of a structural change has occurred at this moment. The image frame index number 500 and the critical node number 50 corresponding to this event are identified. All multidimensional feature parameters at this moment, including angle values, curvature values, and color distribution data, are extracted and encapsulated into independent timestamp event packets to obtain a multi-channel classical clothing design analysis signal set.

[0123] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A classical clothing design analysis system based on image recognition, characterized in that, The system includes: The pattern structure analysis module extracts the continuity index and arrangement density parameters of the boundary between adjacent pattern units based on the local gray-level gradient distribution sequence of classical clothing images, locates abnormal areas of pattern overlap, and generates a pattern structure distribution dataset. The shape and contour modeling module calls the pattern structure distribution dataset, identifies the starting point of the main contour, tracks the position and number of the first curvature change of adjacent contour segments that exceeds the preset benchmark fluctuation threshold, and generates a shape and contour topology map. The color change discrimination module extracts the hue angle difference distribution characteristics of adjacent color blocks based on the shape outline topology map, judges the non-continuous color transition area or local color gamut gradient anomaly point, and generates a color difference detection mark set. The design system modeling module calls the color difference detection mark set, extracts the key turning node numbers and interval pixels of the contour path, analyzes the consistency between the contour direction sequence and the pattern arrangement, and generates a spatial coupling model of classical clothing form and pattern. The design feature output module extracts image numbers and analysis time periods based on the spatial coupling model of classical costume form and pattern, marks style synchronization intervals, and obtains a multi-channel classical costume design analysis signal set. The pattern structure distribution dataset includes pattern overlap anomaly areas, sparse arrangement areas, and boundary break areas. The shape contour topology map includes the main contour starting point, curvature change point, and reference fluctuation boundary point. The color difference detection mark set includes hue angle jump point, color gamut offset break point, and illumination interference sensitive point. The classical clothing shape and pattern space coupling model includes contour direction sequence, pattern arrangement consistency category, and key node interval pixel features. The multi-channel classical clothing design analysis signal set includes analysis image number, potential style time period, and direction synchronization feature identifier.

2. The classical costume design analysis system based on image recognition according to claim 1, characterized in that, The pattern structure analysis module includes: The gradient distribution extraction submodule calculates the first derivative of each pixel in the horizontal and vertical directions based on the gray-level gradient distribution sequence of the local region in the input classical costume image, marks the region where the gradient peak is concentrated, draws the local gradient direction field, evaluates the edge sharpness of the differentiated pattern unit, and generates a gradient direction distribution map. Based on the gradient direction distribution map, the structure comparison submodule compares the boundary continuity index and spatial arrangement density parameter of adjacent pattern units one by one, records the comparison results, marks the location of boundary breakage, counts the number of pattern repetitions per unit area, and generates a pattern structure comparison table. The correlation matrix construction submodule extracts the structural dependence strength of the target pattern unit on the neighboring units in terms of spatial layout based on the pattern structure comparison table, selects high-frequency interaction areas for key modeling, identifies the spatial coupling relationship between pattern units, and generates a pattern structure distribution dataset.

3. The classical costume design analysis system based on image recognition according to claim 2, characterized in that, The shape and outline modeling module includes: The outline starting point recognition submodule calls the pattern structure distribution dataset to detect the gradient convergence center of the main outline of the garment piece at the intersection of the shoulder and neck, marks it as the starting point of the shape modeling, and generates the starting coordinate set of the main outline; The curvature change tracking submodule calculates the curvature value pixel by pixel along the contour direction based on the starting coordinate set of the main contour, records the curvature change position and the corresponding contour segment number when the curvature first exceeds the preset reference fluctuation threshold, and generates a curvature change point sequence. The extension rate normalization submodule evaluates the pixel distance between adjacent contour segments based on the curvature abrupt change point sequence, performs a ratio calculation of the contour segment length to the number of pixels with curvature response delay, and reorganizes the original rate sequence using the formula: ； Normalized into a contour extension rate index, the rate index of the differentiated contour segments is mapped to a unified coordinate system and contour connection lines are drawn to generate a shape contour topology map. in, This represents the rate of contour extension. Representing the The pixel length of the segment outline. This represents the average pixel length of the contour segment. Representing the The number of pixels for the curvature response delay of the contour segment. Representing the The number of pixels for the curvature response delay of the contour segment. The average number of pixels representing curvature response delay. This represents the total number of outline segments.

4. The classical costume design analysis system based on image recognition according to claim 3, characterized in that, The color change discrimination module includes: The hue difference recognition submodule reads the HSV color space coordinates of the center points of adjacent color blocks according to the shape outline topology map, calculates and records the hue angle difference, groups and counts the hue jump frequency by region, and generates a hue angle difference distribution table. The color logic judgment submodule reads the hue angle difference and spatial adjacency relationship of adjacent color blocks based on the hue angle difference distribution table, performs line-by-line raster scanning along the geometric center axis of the clothing image, identifies color block pairs whose hue abrupt changes exceed the preset hue cutoff threshold and marks their numbers, calculates the spatial distance between adjacent abrupt color blocks, and judges the color gamut offset points caused by color logic breakage or lighting interference in combination with the local color gamut distribution pattern, records the abnormal color block numbers and positions, and generates a set of color difference analysis breakpoint coordinates. The anomaly marker generation submodule reads the color block number, image coordinates and corresponding contour position of the breakpoint according to the color difference analysis breakpoint coordinate set, performs a line-by-line comparison between the breakpoint position and the high jump section in the hue angle difference distribution table, extracts the hue angle jump point and color gamut offset breakpoint, performs anomaly type encoding and level label assignment for the breakpoint, and generates a color difference detection marker set.

5. The classical costume design analysis system based on image recognition according to claim 4, characterized in that, The method of statistically analyzing hue jump frequency by region refers to dividing the topological map of the garment's shape outline into several continuous horizontal scanning bands along the main axis of the garment's geometric center, counting the total number of events where the hue angle difference between adjacent color block pairs exceeds a first hue preset threshold, and calculating the ratio of the number to the total number of color blocks in the horizontal scanning band to obtain the hue jump frequency of the region. The method of determining color gamut offset points caused by color logic breaks or lighting interference by combining local color gamut distribution patterns refers to simultaneously calculating the saturation difference and brightness difference between adjacent abrupt color blocks. When the hue angle difference between adjacent color blocks is greater than 60 degrees and the saturation difference is less than 10%, the position is determined to be a color logic break. When the hue angle difference between adjacent color blocks is less than 60 degrees and the brightness difference is greater than 30%, the position is determined to be a color gamut offset point caused by lighting interference.

6. The classical costume design analysis system based on image recognition according to claim 4, characterized in that, The design system modeling module includes: The key node extraction submodule calls the color difference detection flag set, reads the contour node number and pixel raster scan index corresponding to the abnormal color block, arranges the key nodes in the shape contour path order, calculates the pixel interval between adjacent key nodes and stores them in pairs, and generates a key node interval pixel table. The orientation consistency analysis submodule reconstructs the key node sequence of the shape path based on the key node interval pixel table, according to the contour number and pixel raster scan index, identifies the orientation angle changes between consecutive key nodes and marks the turning nodes, numbers the order of turning, counts the distribution deviation of turning positions and structural self-similarity patterns between repetitive pattern units within a single image, analyzes the consistency between the contour orientation sequence and the pattern arrangement, and generates an orientation consistency distribution map. The evolution model construction submodule extracts the key node turning sequence and interval pixel number of the path according to the consistency distribution map, aggregates clothing sample groups according to the degree of consistency and marks the evolution category, and generates a spatial coupling model of classical clothing form and pattern.

7. The classical costume design analysis system based on image recognition according to claim 6, characterized in that, The design feature output module includes: The dynasty matching submodule filters out path sequences that meet the style proximity threshold based on the geometric similarity of the outline direction sequence in the spatial coupling model of classical clothing form and pattern, extracts the image number and analysis time period corresponding to the path, and generates a style interval value set. The synchronization interval marking submodule marks the time intervals of style synchronization based on the image number and analysis time period in the style interval value set, and obtains the design style synchronization interval segment value according to the consistency of the rhythm of form change in the direction sequence. The key event localization submodule extracts the rate of change of direction angle and the ratio of interval pixels based on the image number and analysis time period in the synchronous interval value of the design style, and filters the clothing sample sequence that exceeds the form change threshold, locates the corresponding image sample and key node number, and obtains a multi-channel classical clothing design analysis signal set.

8. The classical costume design analysis system based on image recognition according to claim 7, characterized in that, The time segment for style synchronization refers to reading the contour slope data corresponding to the consecutive image numbers in the style interval value set, calculating the correlation coefficient of the contour slope between adjacent frames, and determining the continuous time range with a correlation coefficient greater than 0.9 and a continuous frame count of more than 10 frames as a segment with consistent shape change rhythm. The start time point and end time point of the segment are recorded as the time segment for style synchronization. The screening of clothing sample sequences that exceed the shape mutation threshold refers to the real-time comparison of the extracted direction angle change rate and the interval pixel ratio. When the direction angle change rate is greater than 30 degrees and the interval pixel ratio changes by more than 20%, the currently processed clothing sample sequence is determined to be a shape mutation sequence.

Images