Research On Key Technologies Of Virtual Dressing System Based On Web

At present, in the process of buying clothes over the web store, potentialbuyers often interrupt the purchase behavior because of unable to estimate fittingeffect accurately. This situation becomes an important factor restricting the businessturnover growth. In recent years, web-based virtual dressing application graduallyincreased. Using virtual dressing, potential buyers can get the intuitiveunderstanding of fitting effect, so as to increase the chance to buy clothes. How toshow the fitting effect to users directly, and enhance the user experience in theprocess of online shopping, is the new research subject in front of scientificresearchers.Web-based virtual dressing involves many techniques, such as computergraphics, computer vision, image processing, image analysis, image understanding,machine learning, artificial intelligence, pattern recognition, web technology, etc.Its research and development is difficulty. According to the current situation of thedevelopment of web technology, the web applications based on3D model havesome bottlenecks in terms of performance. In order to alleviate this problem, theweb-based virtual dressing is defined as virtual dressing based on images in thispaper. There are some key issues in this field, including human detection and poseestimation from a single image, matting and extracting the faces and clothes,optimization and adjustment of the faces, virtual dressing by matching the clothesand bodies. In view of the above problems, this paper studied the key technologiesof human detection, pose estimation, image matting, image deformation, colorfusion, and proposed the corresponding solution.In order to accurately estimate the human pose from a single photo, the paperproposed a scheme of pose estimation based on multiple positioning. The schemefirst uses a algorithm based on improved Harr features cascade to locate frontal faceand shoulders in a single photograph, then uses a hue-based adaptive clusteringalgorithm to determine the position of limbs, and at last uses the prior knowledge ofhuman skeleton to estimate the body posture in the photo. In order to effectivelyextract the head and clothes from the photo, the paper proposed a scheme offoreground extraction based on semi automatic matting. The scheme first setssampling points automatically according to the results of pose estimation, then usesa matting algorithm based on improved GrabCut to extract the head and clothesfrom the photo automatically. In order to obtain satisfactory dressing effect, thepaper uses both a skin color adjusting algorithm based on histogram specification and a algorithm of clothes deformation based on mesh deformation. The skin coloradjusting algorithm can beautify the user’s face by changing skin color. Thealgorithm of clothes deformation deforms the clothes to optimize the effect ofmatching clothes and body during the process of dressing.In order to verify the correctness and effectiveness of the algorithm, this paperdesigned and implemented a web-based virtual dressing system, established theclothing database, and many experiments of virtual dressing proved the desiredeffect.