Research On Image Processing And Size Gaininig Based On The 2d-3d Body Measurement System

This subject is from the project "electronic tailored clothing-electronic made to measure software development and industrial applications", which is from Shanghai Municipal Committee for Economic Development. The main task of this subject is to establish a rapid, accurate and low-cost automatic body measurement system, then together with other sub-projects to accomplish informationization and industrialization of apparel enterprises, including electronic tailored, on-line customers to customize and virtual fitting and so on.The subject uses more advanced image processing algorithms for image processing on the 22 adult male body images and obtains their body size of relevant parts based on the 2D-3D body scanning devices,which is from rapid response research laboratory of garment college.And the processing algorithms include image preprocessing, contour extraction, feature point extraction, and size conversion of relevant parts. Then compare with the data from system and manual measurement, analyze and prove the measurement accuracy of measured size of the relevant parts has reached the demands of clothing production. First of all, on the basis of analyzing the theory and application of several common non-contact measuring system nowadays, this paper determined the technical line and research methods; successfully collected 22 the adult male body front,back and profile images by using the front and profile camera of 2D-3D body scanning devices, at the same time, accurately got 20 related body size by using Martin measurement, and then got the grayscale image of the BMP format which is benefit for computer processing through conversing the acquired color images to grayscale.Secondly, on the treatment of gray-scale normalized, the author using Matlab programming to human body images to conduct gray-scale transformation and histogram equalization, and compared their differences, by analysis got gray-scale transformation is more desirable method to gray-scale normalization in this study; followed analysis the noise inherent types of the camera, and overview several common denoising algorithms, by comparing the differences of human body image after using the mean filtering, median filtering, Wiener filtering and wavelet de-noising treatment, determine the wavelet denoising is the ideal de-noising algorithm in this study; and then conduct sharpened treatment to human body image by gradient method, obtained the human body front,back and profile images which is more obvious in contrast, more clearer and meet the requirments better. Thirdly, discussed the principles and methods of several classical edge detection algorithm, realized five algorithms of the Roberts operator, Sobel operator, Prewitt operator, Laplacian of Gaussian and Canny by using Matlab language to program, selected Prewitt operator as the most ideal edge detection algorithm in this study, and also successfully detected the edge of the human body; and then refined the image by using mathematical morphological algorithm, obtained a single-pixel-wide edge, and achieved a good refinement results; then tracked the human body images by using raster scanning, and obtained the accurate body outside contour image which has smooth edge.Finally, combined the basic structural characteristics of male body,body characteristics and the proportion of the human body, through analysis the corresponding feature points of relevant parts, successfully extracted the relevant feature point main by scanning algorithm, and received the system size of measurement parts by coordinate correction and size conversion, and then calculate the circumference size by using the relationship of width,thickness and circumference; and designed the entire system by using the method of C# program calls Matlab language, and also accounted all of the male body images, accessed the system size of measurement parts; by comparison with manual measurements, it's verified that the accuracy of system measurements basically met the demand of clothing production, and analyzed the error reasons between them, and also pointed out the deficiencies existing in the system and the future research points.The innovation of this paper were:respectively designed three tie belts for chest, waist, hip circumference which enabled men not to be confined by wearing tight-fitting clothes, but also more importantly was marked with a location in the body outline of the image, which can make the identification easier and measurement for length more accurate; the recognition algorithm for feature points was partly optimized to improve the efficiency of its operations, such as converted the completion after separately extracting the tie belts, which shortened the scanning recognition time in multiples.