Research On Key Methods Of Virtual Dressing System

Although the development of apparel e-commerce is rapid in recent years, that users, only based on picture and document information, can not accurately estimate the effort of their own fitting not only decreases the number of online shopping clothing, but also causes a higher rate of return and exchange of clothing, so limiting the development of online shopping garment industry. To address this limitation, a growing number of researchers work on technology related in virtual fitting system,building a virtual fitting system, to provide users with a virtual try on experience.At present, research on the related technology of virtual fitting system has made some achievements, but there are still some issues that need further study. In this paper, the key algorithm of virtual fitting system is explored, mainly on how to solve the problem in the following three aspects: how to generate a garment library used to virtual try on, how to make the head region in virtual try-on image being user head image, and how to virtual try on cloth.In order to better meet the need of generating clothing database in image-based virtual fitting system, the adaptive clothing target matting method is proposed in the paper, which is based on One Cut and shared matting algorithm. First, we utilize segmentation algorithm named as One Cut to segment the cloth image from background. Second, the trimap is automatically generated by mathematical morphology methods. Third, shared matting algorithm is used to obtain the clothing foreground and alpha image used to implement virtual try-on. To make the experience of virtual try-on more realistic, a automatic method that can generate the personalized head image is proposed in the paper, and users can get their own special model image with this method. At first, the face detection algorithm is used to detect human face regions. Then the face feature points in the face area is localized,and the head foreground image is extracted from background image. Finally, calculate the transformation matrix used to transplant the head from the user photo to the body of the model photo, synthetic personalized model image. Meanwhile, in order to make sure that head transplantation of skin is more real and neck migration is seamless,algorithms is utilized to achieve the ultimate exclusive model images, such as skin color transfer and removing the neck area in user image. In this paper, the advantage that both the model image and the image of clothing have basic consistent standing posture is used, so virtual try-on can be realized only by image distortion and image interpolation. With tag information of garment image and model image, you can achieve a key function to try on, and also adjust the effect of the virtual try-on by manual interaction. Meanwhile, the design and implementation of the virtual fittingsystem applications is completed by using above methods. Its main function includes:extracting apparel target foreground image based on the original image of clothing,then establishing clothing foreground library; according to the model image and the user’s head image, generating personalized head image; selecting apparel image in clothing foreground library to virtual try on.A large number of experiments show: the proposed method, the adaptive clothing target matting method, not only meets the real-time requirements, and in the case of less interaction the result of garment matting is better. The method of personalizing header image does not require user interaction, but can achieve the consequent of seamless head transplantation quickly and efficiently. With using the classic image warping and interpolation algorithms, virtual try-on achieve a strong realistic effect of trying on the cloth.