A human body image size extraction method based on wavelet algorithm

Abstract

The application discloses a human body image size extraction method based on a wavelet algorithm, and relates to the technical field of human body measurement. The method comprises the following steps: collecting images, unifying specifications, preprocessing, estimating postures, segmenting, extracting contours, extracting feature points, and calculating sizes. The front, side and back images of a standard posture are collected, and after the specifications are unified and enhanced, weighted average gray conversion, wavelet joint denoising and Canny edge detection preprocessing are performed. The human body region is extracted through posture estimation and adaptive segmentation, and the feature points are positioned by combining human body proportions and dynamic programming. Two-dimensional sizes are obtained by using the height proportion method, and three-dimensional sizes are obtained by fitting a double-ellipse model. The method is convenient to operate, low in cost, high in precision, and can meet the human body size requirements of personalized clothing customization.

Description

Technical Field

[0001] This invention relates to the field of human body size measurement technology, specifically to a method for extracting human body image size based on wavelet algorithm. It is applicable to scenarios such as personalized clothing customization, human body data collection and analysis, and e-commerce clothing recommendation, and can especially meet the need for rapid acquisition of accurate human body size in large-scale customized production. Background Technology

[0002] With the rapid development of e-commerce and big data, the problem of "information overload" in the apparel consumption sector has become prominent, leading to persistently high return rates. Consumers' demand for personalized, well-fitting clothing is increasingly evident, making the transformation of the apparel industry towards personalized services an inevitable trend. Accurate anthropometric data is the core engine driving this transformation. Apparel sizing optimization is a systematic project encompassing consumer experience, industry efficiency, and social inclusion. Traditional sizing standards have poor adaptability and struggle to meet the body shape needs of different consumers. Building a new sizing system to respond to market demands is a practical necessity, and big data technology provides strong support for sizing optimization, making it feasible. Accurate anthropometric data can optimize clothing pattern design, construct scientifically sound sizing standards, effectively reduce clothing return rates, and simultaneously improve product quality and overall industry efficiency.

[0003] Currently, non-contact 3D automatic human body measurement has become a key research focus and development direction. Researchers are using portable cameras, mobile phone photography and 3D reconstruction, computer vision technology, and other methods to achieve 3D reconstruction. However, single-view 2D images cannot obtain the 3D structural information of the target, requiring multi-view 2D images to support high-precision reconstruction. Therefore, the choice of equipment and the 3D reconstruction method of multi-view image sequences are crucial to improving the speed and accuracy of human body measurement. Existing related technical solutions include using two cameras to acquire frontal and side view images of the human body, processing them with algorithms to obtain the human body contour, and reconstructing the height, length, and girth dimensions. However, due to the small sample size, there are certain errors when calculating the girth dimensions using human body size regression models. Another example is the development of a mobile phone-based human body measurement platform using 2D images captured by a mobile phone, which is widely used by ordinary users to estimate key body parameters, but the platform's functionality is relatively basic. A human body size measurement system developed based on an Android mobile terminal, although embedded in a Chinese online clothing shopping platform to provide body size data recommendations for clothing buyers, does not consider the elasticity and thickness of clothing fabrics, resulting in an incomplete classification of human body shapes. A third method involves using a 3D human body scanner to collect human body data and obtain a parametric template model of a cheongsam, but 3D human body scanners are expensive and lack portability, hindering widespread application. Image size measurement, compared to traditional measurement, has the advantages of being non-contact and highly accurate, and after algorithm optimization, it can adapt to the needs of modern industrial high-efficiency production. However, in practical applications, human body image measurement suffers from large errors and insufficient applicability due to factors such as clothing, shooting conditions, and image noise. Furthermore, the image segmentation and key point detection steps in the process of acquiring human body dimensions can lead to error accumulation, affecting the accuracy of the final measurement results.

[0004] Therefore, there is an urgent need to develop a more efficient, accurate, low-cost, and portable method for extracting human body image dimensions to meet the high-precision requirements for human body size data in fields such as personalized clothing customization, and to provide technical support for size optimization and personalized service transformation in the clothing industry. Summary of the Invention

[0006] The purpose of this invention is to overcome the problems of large errors, insufficient applicability, high cost, complex operation, and poor portability in existing human body size measurement technologies. Specifically, these include: traditional manual measurement is inefficient and subjectivity leads to large errors; traditional 3D scanning instruments are expensive and inconvenient to use, making them difficult to popularize; existing image measurement methods are affected by shooting conditions, image noise, etc., resulting in significant error accumulation and insufficient size calculation accuracy. This invention provides a human body image size extraction method based on wavelet algorithm.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution: This invention discloses a method for extracting the size of human images based on wavelet algorithm, comprising the following steps: Step 1) Image acquisition: Use a mobile phone to take images of the front, side and back of the human body. Ensure that the light is uniform and bright when taking the picture. The subject should wear fitted clothing, stand with both feet in the designated position, keep the body straight, look straight ahead, shoulders the same width, arms naturally open at 15-45 degrees, palms facing inward and slightly clenched into fists. The shooting distance is one meter, and the height is fixed by a mobile phone stand. Step 2) Image standardization: Standardize the acquired front, side and back images of the human body to ensure consistent image size and resolution; Step 3) Image Enhancement: Perform histogram equalization on the standardized images to enhance image contrast; Step 4) Image preprocessing: The enhanced image is subjected to grayscale conversion, denoising, and edge detection in sequence. The grayscale conversion adopts the weighted average method, the denoising adopts the joint optimization denoising method of nonlocal mean and wavelet transform, and the edge detection adopts the Canny edge detection algorithm. Step 5) Human pose estimation: Perform human pose analysis based on the preprocessed image to determine the human pose state in the image; Step 6) Adaptive segmentation: Based on the human pose estimation results, the image is adaptively segmented to separate the human body region from the background region; Step 7) Human contour extraction: Extract the human contour from the segmented human body region, generate a binary image of the human contour, obtain the coordinates of the human contour and draw a rectangle. Step 8) Human Feature Point Extraction: Based on human contour coordinates and rectangular box information, a key point detection algorithm based on human proportion is adopted. Local extrema are searched by sliding window method, and feature point path is dynamically planned and optimized to achieve localization. For front and back images, key points of the top of the head, shoulders and feet are extracted, and for side images, key points of the top of the head, shoulders, chest, waist, hips and feet are extracted. Step 9) Human body size calculation: The user submits their height to generate known actual dimensions. Based on the extracted feature points, a two-dimensional height ratio method is used. The human body ratio coefficient is obtained by using the Euclidean distance between the user's input height and the outline border. The two-dimensional size data is obtained by combining the pixel information of each part. Finally, the three-dimensional size data is obtained by fitting a three-dimensional double ellipse model and combining it with linear correction.

[0008] Preferably, in step 4), the denoising process uses Haar or Db4 wavelet basis for 3-level decomposition, performs soft thresholding on the high-frequency subband coefficients, and then performs wavelet reconstruction to effectively remove Gaussian noise or random noise while preserving the structural edge details of the human body contour, joints, etc.

[0009] Preferably, in step 4), the Canny edge detection algorithm enhances the continuity of image edges and accurately extracts human body edge contours through Gaussian filtering to smooth the image, calculating gradient magnitude and direction, non-maximum suppression, double threshold detection, and edge connection steps.

[0010] Preferably, the two-dimensional dimensional data in step 9) includes height, neck length, chest width, leg length, arm length, and shoulder width.

[0011] Preferably, the three-dimensional dimensional data in step 9) includes neck circumference, chest circumference, waist circumference, and hip circumference; Calculation formula: Dbust=16.202+2.13×Tbust+0.93×Wbust+0.172×W−0.761×Tshoulder; where Dbust is the output chest circumference information, Tbust is the torso length measurement parameter corresponding to the chest circumference, Wbust is the width measurement parameter corresponding to the chest circumference, Tshoulder is the torso feature parameter corresponding to the shoulder, and W is the weight information.

[0012] Preferably, when the double ellipse model algorithm fits the three-dimensional dimensional data, it first calculates the two-dimensional dimensional data of chest width, chest thickness, waist width, and waist thickness. Formulas for shoulder width related parameters: in, The vertical coordinate of the feature point on the right shoulder is shown. The vertical coordinate of the feature point on the left shoulder; Formulas for characteristic parameters of human body parts: ;in, These are the ordinate mapping values ​​of the feature points corresponding to different parts of the human body; Formula for calculating three-dimensional dimensions: L=π(bi−b1)+4ai−2(bi+b1); Where L represents the three-dimensional scale information of the human waist and hips, and ai represents the semi-major axis parameter of the ellipse.

[0013] Preferably, the shooting device in step 1) is an Apple 13 Pro Max mobile phone, which meets the requirements for image clarity and stability.

[0014] Preferably, in step 9), a multi-scale wavelet fusion strategy is adopted in the calculation of human body size, and high-frequency information from different decomposition layers is dynamically weighted and fused. At the same time, feature search with anatomical constraints and a fast calibration and compensation algorithm based on projective geometry are combined.

[0015] Preferably, the anatomically constrained feature search transforms human proportion features into an optimization objective function, where the waist-to-hip ratio is set as a constraint condition of 0.7.

[0016] Preferably, the fast calibration compensation algorithm supports dynamic adjustment of the shooting device's pose and achieves automatic calibration based on the principle of projective geometry, ensuring measurement accuracy.

[0017] Beneficial effects: Accuracy is ensured through multi-stage technical optimization. Compared to traditional denoising techniques, the non-local mean and wavelet transform combined denoising method better preserves human body edge details. The Canny edge detection algorithm enhances edge continuity. Feature point extraction combined with multiple optimization strategies achieves precise positioning. 3D size calculation employs a double-ellipse model fitting and linear correction, along with multiple optimization techniques. The final measurement maximum error is far less than the maximum sample error required for clothing customization human body measurement, ensuring reliable measurement results. Integrating steps such as image specification unification, human pose estimation, and adaptive segmentation effectively eliminates the influence of image specification differences, human pose variations, and background interference. It can adapt to different shooting scenarios and human poses, while also supporting portable mobile phone shooting without the need for professional measurement. The measurement environment is diverse, with a wide range of applications. It can be captured using an iPhone 13 Pro Max, eliminating the need for expensive professional measuring instruments. The shooting process is simple and easy to operate, allowing ordinary users to complete the task. Data processing is highly automated, requiring no complex manual intervention, reducing measurement costs and operational barriers, and facilitating widespread adoption. The methodology, from image acquisition to size calculation, is tightly integrated and logically clear, forming a complete technical system that effectively reduces error accumulation in intermediate stages, improving overall measurement stability and consistency. It provides accurate human body size data for personalized clothing customization, helps build scientific sizing standards, reduces clothing return rates, improves product quality and industry efficiency, and provides technical support for the digital transformation of the clothing industry, possessing significant industrial value. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the process of this invention.

[0019] Figure 2 This is the grayscale image and enhanced image of the present invention.

[0020] Figure 3 This is a comparison chart of the thermal differences between Gaussian filtering and wavelet NLM denoising in this invention.

[0021] Figure 4 These are comparison images of edge detection in two different denoised images according to the present invention.

[0022] Figure 5 This is the contour map of the front, side and back edge detection of the wavelet joint denoising method of the present invention.

[0023] Figure 6 This is the human body frontal feature point extraction image of the present invention.

[0024] Figure 7 This invention uses fitting to obtain three-dimensional dimensional data maps. Detailed Implementation

[0025] The following will refer to the accompanying drawings in the embodiments of the present invention. Figures 1-7 The technical solutions in the embodiments of the present invention are clearly and completely described herein. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention aims to provide a method for extracting human body image dimensions based on wavelet algorithms. By designing a complete measurement process, optimizing image preprocessing algorithms, accurately extracting human body feature points, improving the dimension calculation model, and incorporating multiple optimization techniques, it achieves fast, accurate, low-cost, non-contact, and portable human body dimension extraction. This meets the high-precision requirements for human body dimension data in scenarios such as personalized clothing customization and e-commerce clothing recommendations, provides solid technical support for size optimization and personalized service transformation in the clothing industry, improves the efficiency of large-scale customized production, and promotes the digital and intelligent development of the clothing industry.

[0027] Invention content / principle: This invention discloses a method for extracting human body image dimensions based on wavelet algorithm. It achieves accurate and efficient human body dimension extraction through a complete nine-step process, with each step progressively and synergistically working together. Specific technical details are as follows: Image Acquisition: If an Apple iPhone 13 Pro Max is used as the shooting device, its high pixel count and stable shooting performance will meet the requirements for image clarity and stability. The shooting environment should be bright and evenly lit, avoiding shadows or reflections. The subject should wear fitted clothing to eliminate any obstruction of the body's contours. The required posture is as follows: feet shoulder-width apart, body upright, eyes looking straight ahead, arms naturally extended at a 15-45 degree angle, palms slightly clenched inwards, ensuring a standardized and uniform posture. The shooting distance should be strictly controlled at one meter, using a phone stand to fix the height for shooting. Three images should be captured: front, side, and back views, to minimize errors caused by shooting instability.

[0028] Image standardization: The three human body images (front, side, and back) are standardized by adjusting all images to the same size and resolution. This eliminates the interference of image size differences from different shooting scenarios on subsequent processing and ensures the consistency and accuracy of data processing.

[0029] Image enhancement: Histogram equalization is used to process images after they have been standardized. By adjusting the distribution range of gray values, the overall contrast of the image is enhanced, making the boundary between the human body area and the background area clearer, thus laying a high-quality foundation for subsequent image preprocessing and feature extraction.

[0030] Image preprocessing: Grayscale conversion uses a weighted average method to convert the color image into a grayscale image. By assigning appropriate weights to the RGB channels, the brightness information and key details of the image are preserved to the greatest extent. Denoising: A joint optimization denoising method of nonlocal mean and wavelet transform is used. The grayscale image is decomposed into a three-level decomposition using Haar or Db4 wavelet basis, dividing the image into a low-frequency sub-band containing the main information and a high-frequency sub-band containing noise and edge details. Soft thresholding is applied to the coefficients of the high-frequency sub-band to remove noise components, followed by wavelet reconstruction. This method can effectively remove Gaussian noise or random noise while fully preserving the edge details of human contours, joints, and other structures. Edge detection: The Canny edge detection algorithm is used. Through five steps—Gaussian filtering to smooth the image, calculating gradient magnitude and direction, non-maximum suppression, double threshold detection, and edge connection—the continuity of image edges is enhanced, and human edge contours are accurately extracted, providing a clear contour foundation for subsequent processing.

[0031] Specifically, such as Figure 3 As shown, a joint optimization denoising method using nonlocal mean and wavelet transform was employed, which performed better in edge preservation compared to Gaussian filtering. A denoised image was obtained by using a Haar or Db4 wavelet basis for three-level decomposition, soft thresholding of high-frequency subband coefficients, and wavelet reconstruction, which preserved edge information better than median filtering. In Matlab, the original reference image was visualized and compared with the Gaussian-filtered denoised image and the wavelet NLM joint denoised image to be evaluated, resulting in a difference heatmap (e.g.,...). Figure 3 As shown in the figure, the image obtained by Gaussian filtering denoising has a PSNR of 37.20dB and an SSIM of 0.9456. The image obtained by wavelet NLM joint denoising has a PSNR of 69.90dB and an SSIM of 1.00, indicating that wavelet NLM has a better denoising effect.

[0032] Human pose estimation: Based on the preprocessed image, computer vision technology is used to analyze the human pose state, including key information such as body tilt angle and limb position, to provide pose reference for subsequent adaptive segmentation and feature point extraction, ensuring that the segmentation and extraction process can adapt to different human poses and improve the applicability of the method.

[0033] Adaptive segmentation: Based on the human pose estimation results, combined with information such as image grayscale values ​​and texture features, an adaptive segmentation algorithm is used to automatically separate the human body region from the background region, eliminate background interference, and obtain a clean human body region image, providing a precise processing object for human contour extraction.

[0034] Human contour extraction: From the adaptively segmented human body region image, the complete human contour is extracted and a binary image of the human contour is generated (containing only black and white pixels, with white representing the human contour and black representing the background). The coordinate information of the human contour is obtained through image processing algorithms, and a rectangular box surrounding the human contour is drawn based on the coordinate information, providing a spatial reference for feature point extraction and size calculation.

[0035] Human Feature Point Extraction: Based on human contour coordinates and bounding box information, a key point detection algorithm based on human proportions is employed, combined with the sliding window method and dynamic programming optimization to achieve accurate feature point localization. For frontal and back images, key points are extracted from the top of the head, shoulders, and soles of the feet; for side images, key points are extracted from the top of the head, shoulders, chest, waist, hips, and soles of the feet. Local extrema are searched using the sliding window method to initially determine candidate locations for feature points. Then, dynamic programming is used to optimize the feature point paths and eliminate erroneous candidate points to ensure accurate feature point localization.

[0036] Human body size calculation: Two-dimensional size calculation. The user submits their height to generate known actual size. The two-dimensional height ratio method is used to calculate the human body ratio coefficient by using the Euclidean distance between the user's input height and the outline border. Combined with the pixel information of each part, two-dimensional size data such as height, neck length, chest width, leg length, arm length, and shoulder width are obtained.

[0037] The relevant formulas for calculating the scale factor in the height-ratio method are as follows: ; rs is the scale factor; H is the actual height of the human body (the user's own height data); n is the number of feature points involved in the calculation (the total number of key points selected on the human body contour); xi is the actual ordinate of the i-th feature point (the feature point position parameter based on the actual size of the human body); yi is the pixel ordinate of the i-th feature point (the pixel coordinates of the feature point extracted from the acquired image).

[0038] Three-dimensional dimensional calculations are performed, with 3D dimensional data including neck circumference, chest circumference, waist circumference, and hip circumference obtained through a 3D double-ellipse model fitting combined with linear correction. Before fitting, two-dimensional dimensional data such as chest width, chest thickness, waist width, and waist thickness are calculated. Then, fitting is performed based on the double-ellipse model algorithm, specifically calculated using the following formula: Chest circumference calculation formula: Dbust=16.202+2.13×Tbust+0.93×Wbust+0.172×W−0.761×Tshoulder, where Dbust is the output chest circumference information, Tbust is the torso length measurement parameter corresponding to the chest circumference, Wbust is the width measurement parameter corresponding to the chest circumference, Tshoulder is the torso feature parameter corresponding to the shoulder, and W is the weight information.

[0039] Formulas for shoulder width related parameters: in, The vertical coordinate of the feature point on the right shoulder is shown. The vertical coordinate of the feature point on the left shoulder; Formulas for characteristic parameters of human body parts: ;in, This represents the ordinate mapping value of the feature points corresponding to the human body parts.

[0040] Formula for calculating the three-dimensional dimensions of the waist and hips: L=π(bi−b1)+4ai−2(bi+b1), where L is the three-dimensional scale information of the human waist and hips, and ai is the semi-major axis parameter of the ellipse.

[0041] The size calculation process incorporates a multi-scale wavelet fusion strategy, dynamically weighting and fusing high-frequency information from different decomposition layers to suppress noise while preserving detailed features. Combined with anatomically constrained feature search, the human body proportion features (with waist-to-hip ratio set as a constraint of 0.7) are transformed into an optimization objective function to improve positioning robustness. A fast calibration compensation algorithm based on projective geometry is adopted to support dynamic adjustment of the shooting device pose, achieve automatic calibration, and ensure measurement accuracy.

[0042] Example 1: Using a random sampling method, nulliparous women aged 18-30 in the Jiangsu and Zhejiang regions were selected as the experimental sample. A total of 30 subjects were selected. This group has a relatively stable body shape, which can reduce the interference of body shape changes caused by factors such as childbirth on the measurement results, and ensure the reliability and representativeness of the experimental data.

[0043] Measurement tool preparation: The shooting device is an Apple 13 Pro Max mobile phone, equipped with an adjustable height phone stand; the manual measurement equipment is a soft ruler with an accuracy of 0.1cm and a human height measuring instrument, which are used to verify the accuracy of the image measurement results; the measurement system can be equipped with dedicated data processing software based on Matlab and OpenCV, which supports functions such as image preprocessing, feature point extraction, and size calculation.

[0044] Standard Basis: Strictly follow the national standards GB / T16160-2017 "Definition and Method of Human Body Measurement for Clothing" and GB / T38131-2019 "Method for Obtaining Reference Points for Human Body Measurement for Clothing" as well as the actual needs of clothing production, and determine 10 measurement points: height, neck circumference, chest circumference, waist circumference, hip circumference, neck length, front chest width, leg length, shoulder width, and total arm length. Specific implementation steps

[0045] 1. Image acquisition implementation: Shooting environment setup: Choose an indoor location with even and bright lighting, avoiding direct sunlight and shadows. The background should be a solid color without complex decorations to reduce background interference.

[0046] Preparation of the test subject: The test subject should wear fitted tight-fitting clothes, remove any excess jewelry, stand in the designated position with feet together, body straight, eyes looking straight ahead, shoulders shoulder-width apart, arms naturally open at 15-45 degrees, palms facing inward and slightly clenched into fists, and maintain a stable posture.

[0047] Shooting procedure: Fix the iPhone 13 Pro Max on the stand, adjust the height of the stand so that the phone lens is roughly level with the subject's shoulder, maintain a shooting distance of one meter, and take pictures of the subject's front, side and back. Take three pictures of each image, and select the clearest and most symmetrical image for subsequent processing.

[0048] 2. Unified Image Specification Implementation: Using data processing software, all acquired images are uniformly adjusted to a resolution of 1920×1080 pixels, and the image format is uniformly set to JPEG format to ensure that all images are consistent in specifications and eliminate differences in image size and resolution caused by different shooting angles and number of shots.

[0049] 3. Image Enhancement Implementation: The image after standardization is processed by the histogram equalization function in the software (such as the histeq function in Matlab), which automatically adjusts the gray-level histogram distribution of the image, so that the gray values ​​are evenly distributed in the range of 0-255 gray levels, thereby enhancing the image contrast and making the human body contours more distinct from the background.

[0050] 4. Image preprocessing implementation: Grayscale conversion: A weighted average method is used, and the color image is converted into a grayscale image by the formula Gray=0.299R+0.587G+0.114B (where R, G, and B are the pixel values ​​of the red, green, and blue channels of the color image), while retaining the image brightness information.

[0051] Denoising: In the software, the Haar wavelet basis is used to perform a 3-level wavelet decomposition on the grayscale image, resulting in one low-frequency subband coefficient and three high-frequency subband coefficients. Soft thresholding is applied to the high-frequency subband coefficients; the threshold calculation formula is as follows: (σ is the noise standard deviation, N is the total number of image pixels), set the high-frequency coefficients whose absolute value is less than the threshold to zero, and subtract the threshold from the high-frequency coefficients whose absolute value is greater than the threshold; finally, perform wavelet reconstruction on the processed wavelet coefficients to obtain the denoised image.

[0052] Edge detection: The Canny edge detection algorithm is adopted, with a low threshold of 0.1 and a high threshold of 0.3. The algorithm is implemented through the edge function in the software. The algorithm first performs Gaussian filtering to smooth the image, then calculates the gradient magnitude and direction, refines the edges through non-maximum suppression, and finally obtains a continuous and clear human body edge contour through double threshold detection and edge connection.

[0053] 5. Human pose estimation implementation: For example, based on the human pose estimation algorithm in the OpenCV open source library, the image after edge detection is processed to detect the positions of key joints such as shoulders, elbows, hips, and knees in the image, calculate the angular relationship between joints, determine the human body tilt angle, arm opening angle and other pose information, and output the pose information as the input parameters for subsequent adaptive segmentation.

[0054] 6. Adaptive Segmentation Implementation: Based on the human pose estimation results, a threshold-based adaptive segmentation algorithm is adopted. The optimal segmentation threshold is determined through software iterative calculation to maximize the inter-class variance between the segmented human and background regions. The image is then divided into human and background regions based on the optimal threshold. The background region is filled with black to obtain a clean image containing only the human region. Human Contour Extraction Implementation: For example, the bwboundaries function is used to extract the contours of the adaptively segmented human region image, generating a binary image of the human contour. The (x, y) coordinates of each pixel on the contour are obtained through a coordinate extraction function. Based on the coordinate information, the minimum bounding rectangle of the contour is calculated, and a rectangular box enclosing the human contour is drawn. The height and width pixel values ​​of the rectangle are recorded.

[0055] 7. Human Feature Point Extraction Implementation: Set the proportion range of each feature point: for example, the top of the head is located in the range of 0-5% of the height of the rectangle, the shoulder key point is located in the range of 45%-55%, the chest key point is located in the range of 55%-65%, the waist key point is located in the range of 65%-75%, the hip key point is located in the range of 75%-85%, and the foot point is located in the range of 95%-100%.

[0056] Sliding window search: Set a 3×3 pixel sliding window, slide it within the proportional range corresponding to each feature point, calculate the extreme values ​​of pixel grayscale values ​​within the window, and preliminarily determine the candidate positions of feature points.

[0057] Dynamic programming optimization: Using the distance and angle constraints between adjacent feature points as the objective function, candidate positions are optimized and filtered, erroneous candidate points that deviate from the human body structure are eliminated, the final feature point positions are determined and the coordinates are recorded.

[0058] 9. Implementation of Human Body Size Calculation: Two-dimensional dimension calculation: For example, the actual height (e.g., 168cm); calculate the height of the rectangle in pixels (e.g., 1000 pixels), and get the human body proportion coefficient = actualHeight / height of the rectangle in pixels = 168cm / 1000 pixels = 0.168cm / pixel; measure the pixel length of each two-dimensional dimension part on the outline (e.g., shoulder width in pixels, 265 pixels), then the actual shoulder width = 265 pixels × 0.168cm / pixel ≈ 44.5cm. Similarly, calculate the two-dimensional dimension data such as height, neck length, chest width, leg length, and arm length.

[0059] 3D Dimension Calculation: Basic Parameter Calculation: Calculate 2D dimensional data such as chest width (pixel length between feature points on both sides of the chest × proportional coefficient), chest thickness (estimated depth of chest feature points from the front and side), waist width (pixel length between feature points on both sides of the waist × proportional coefficient), and waist thickness (estimated depth of waist feature points from the front and side) based on feature point coordinates.

[0060] Chest circumference calculation: It is calculated using the formula Dbust=16.202+2.13×Tbust+0.93×Wbust+0.172×W−0.761×Tshoulder, where Tbust is the pixel distance between the shoulder and chest feature points × the scaling factor, Wbust is the chest width, W is the user-input weight (e.g., 55kg), and Tshoulder is the average curvature of the shoulder contour (calculated from the coordinates of the feature points).

[0061] Shoulder width related parameters calculation: through formula Calculate and substitute the ordinate values ​​of the feature points on the right and left shoulders (e.g., ...). =500 pixels, =235 pixels, then b1=(500−235) / 2=132.5 pixels).

[0062] Calculation of characteristic parameters of human body parts: through formulas Calculate and substitute the ordinate mapping values ​​of the corresponding human body feature points (e.g., K(yj)=400, K(yi)=300, then bi=(400−300) / 2=50). Waist and hip circumference calculation: Calculated using the formula L=π(bi−b1)+4ai−2(bi+b1), where ai is the semi-major axis parameter of the ellipse (waist width / 2 or hip width / 2). Substitute the relevant parameters to obtain the three-dimensional scale information of the waist and hips, and then perform linear correction (waist correction coefficient 0.98, hip correction coefficient 0.97) to obtain the final waist and hip circumference data.

[0063] During the size calculation process, a multi-scale wavelet fusion strategy, anatomical constraint feature search (with waist-to-hip ratio set at 0.7 as the constraint), and a fast calibration compensation algorithm can be used to dynamically adjust parameters and ensure measurement accuracy.

[0064] 10. Measurement system stability verification The experiment used the method described in this paper to measure the dimensions of 10 mannequins and 20 subjects of different ages and genders, all wearing close-fitting clothing. The experimental results and the results of manual measurement are shown in Table 1.

[0065] Table 1 Comparison of errors in partial data from manual and programmed measurements. body parts Manual measurement value / cm Image measurement value / cm Absolute error / mm relative error / % height 168 168.5 0.5 0.29 Neck length 7 7.2 0.2 0.28 Neck circumference 35 34.4 0.6 1.15 Wide front chest 32 31.1 0.9 0.28 chest circumference 89.5 90.1 0.6 0.67 waistline 71.5 73.5 2 0.95 Hip circumference 96 99.2 3.2 0.30 Leg length 90 90.7 0.7 0.77 shoulder width 44 44.5 0.5 2.02 Full arm length 62.5 64.3 1.8 0.78 Error assessment: Considering the limitations of human and material resources, key parts of the human body were selected for analysis of absolute and relative errors. Evaluation indicators: 1) Absolute error: |Experimental value - Manually measured value|; 2) Relative error: Absolute error / manual measurement value × 100%; As shown in Table 1, the error between the measured values ​​obtained by the method of this invention and the manual measurement values ​​is small, with no significant difference between the two. For clothing customization, the maximum sample error in human anatomy measurement should not exceed 5 cm, and this sample's maximum error is within 5 cm. However, manual measurement is difficult to determine the narrowest point of the waist, leading to large measurement errors. The hip circumference is a sensitive area of ​​the human body, and manual measurement also introduces errors. The error between the measured values ​​obtained by the method of this invention and the manual measurement values ​​is small, and the measurement data obtained by the method of this invention is relatively stable. Considering the experimental conditions and objective factors such as measurement errors, the measurement results obtained by the method of this invention are basically accurate, meeting the expected requirements of this experiment. This proves that the non-contact human parameter measurement proposed in this paper meets the accuracy requirements and demonstrates that the research on three-dimensional human anatomy measurement based on multi-view reconstruction is feasible and effective.

[0066] 11. Result Validation and Output The two-dimensional and three-dimensional dimensional data obtained by image measurement are compared with manual measurement data to calculate absolute and relative errors and verify measurement accuracy. The measurement method of this invention can integrate all dimensional data and output 10 human body dimensional measurement results, including height, neck length, neck circumference, front chest width, chest circumference, waist circumference, hip circumference, leg length, shoulder width, and total arm length, for use by users or clothing customization systems.

[0067] Finally, it should be noted that the present invention is not limited to the above embodiments, and many variations are possible. All variations that can be directly derived or conceived by those skilled in the art from the disclosure of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for extracting the size of human images based on wavelet algorithm, characterized in that, Includes the following steps: Step 1) Image acquisition: Take images of the front, side and back of the human body. Ensure that the lighting is uniform and bright when taking pictures. The subject should wear fitted clothing, stand with both feet in the designated position, keep the body straight, look straight ahead, shoulders the same width, arms naturally open at 15-45 degrees, palms facing inward and slightly clenched into fists. The shooting distance is one meter. Step 2) Image standardization: Standardize the acquired front, side and back images of the human body to ensure consistent image size and resolution; Step 3) Image Enhancement: Perform histogram equalization on the standardized images to enhance image contrast; Step 4) Image preprocessing: The enhanced image is subjected to grayscale conversion, denoising, and edge detection in sequence. The grayscale conversion adopts the weighted average method, the denoising adopts the joint optimization denoising method of nonlocal mean and wavelet transform, and the edge detection adopts the Canny edge detection algorithm. Step 5) Human pose estimation: Perform human pose analysis based on the preprocessed image to determine the human pose state in the image; Step 6) Adaptive segmentation: Based on the human pose estimation results, the image is adaptively segmented to separate the human body region from the background region; Step 7) Human contour extraction: Extract the human contour from the segmented human body region, generate a binary image of the human contour, obtain the coordinates of the human contour and draw a rectangle. Step 8) Human Feature Point Extraction: Based on human contour coordinates and rectangular box information, a key point detection algorithm based on human proportion is adopted. Local extrema are searched by sliding window method, and feature point path is dynamically planned and optimized to achieve localization. For front and back images, key points of the top of the head, shoulders and feet are extracted, and for side images, key points of the top of the head, shoulders, chest, waist, hips and feet are extracted. Step 9) Human body size calculation: The user submits their height to generate known actual dimensions. Based on the extracted feature points, a two-dimensional height ratio method is used. The human body ratio coefficient is obtained by using the Euclidean distance between the user's input height and the outline border. The two-dimensional size data is obtained by combining the pixel information of each part. Finally, the three-dimensional size data is obtained by fitting a three-dimensional double ellipse model and combining it with linear correction.

2. The method for extracting human image size based on wavelet algorithm according to claim 1, characterized in that, In step 4), the noise reduction process uses Haar or Db4 wavelet basis for 3-level decomposition. After soft thresholding of the high-frequency subband coefficients, wavelet reconstruction is performed to effectively remove Gaussian noise or random noise while preserving the structural edge details of the human body contour, joints, etc.

3. The method for extracting human image size based on wavelet algorithm according to claim 1, characterized in that, In step 4), the Canny edge detection algorithm enhances the continuity of image edges and accurately extracts human body edge contours through Gaussian filtering to smooth the image, calculating gradient magnitude and direction, non-maximum suppression, double threshold detection, and edge connection steps.

4. The method for extracting human body image size based on wavelet algorithm according to claim 1, characterized in that, The two-dimensional dimensional data in step 9) includes height, neck length, chest width, leg length, arm length, and shoulder width.

5. The method for extracting human body image size based on wavelet algorithm according to claim 1, characterized in that, Step 9) includes three-dimensional dimensional data such as neck circumference, chest circumference, waist circumference, and hip circumference; Calculation formula: Dbust=16.202+2.13×Tbust+0.93×Wbust+0.172×W−0.761×Tshoulder; where Dbust is the output chest circumference information, Tbust is the torso length measurement parameter corresponding to the chest circumference, Wbust is the width measurement parameter corresponding to the chest circumference, Tshoulder is the torso feature parameter corresponding to the shoulder, and W is the weight information.

6. The method for extracting human image size based on wavelet algorithm according to claim 5, characterized in that, When fitting three-dimensional dimensional data using the double ellipse model algorithm, two-dimensional dimensional data of chest width, chest thickness, waist width, and waist thickness are first calculated. Formulas for shoulder width related parameters: in, The vertical coordinate of the feature point on the right shoulder is shown. The vertical coordinate of the feature point on the left shoulder; Formulas for characteristic parameters of human body parts: ;in, These are the ordinate mapping values ​​of the feature points corresponding to different parts of the human body; Formula for calculating three-dimensional dimensions: L=π(bi−b1)+4ai−2(bi+b1); Where L represents the three-dimensional scale information of the human waist and hips, and ai represents the semi-major axis parameter of the ellipse.

7. The method for extracting human image size based on wavelet algorithm according to claim 1, characterized in that, In step 1), the shooting device was an Apple 13 Pro Max phone, which met the requirements for image clarity and stability.

8. The method for extracting human body image size based on wavelet algorithm according to claim 1, characterized in that, In step 9), a multi-scale wavelet fusion strategy is adopted in the calculation of human body size. The high-frequency information of different decomposition layers is dynamically weighted and fused. At the same time, it combines feature search with anatomical constraints and a fast calibration and compensation algorithm based on projective geometry.

9. A method for extracting human image size based on wavelet algorithm according to claim 8, characterized in that, The feature search with anatomical constraints transforms human proportion features into an optimization objective function, where the waist-to-hip ratio is set as a constraint of 0.

7.

10. A method for extracting human image size based on wavelet algorithm according to claim 8 or 9, characterized in that, The fast calibration compensation algorithm supports dynamic adjustment of the shooting device's pose and achieves automatic calibration based on the principle of projective geometry, ensuring measurement accuracy.

Images