Research On Image Deformation Methods For Two-dimensional Virtual Try-on

As the earliest category involved in e-commerce,clothing has become the largest and most mature industry.Online clothing sales have many advantages that traditional modes do not have,allowing users to fully enjoy the fun and interactive experience of online shopping,with huge market value and economic benefits.However,the introduction of texts and pictures alone does not provide users with a good shopping experience,which restricts the further development of online clothing sales.In recent years,researchers have put forward many methods for two-dimensional and three-dimensional virtual try-on.Under the background of virtual try-on under network environment,three-dimensional virtual try-in has the disadvantages of high modeling complexity,long processing time,and high production cost.And limitations,can not really be applied to clothing online sales.However,the two-dimensional virtual try-in has a single try-on effect,and can only be aimed at a fixed attitude or a specific user,and cannot meet the increasing demand for virtual try-on.Therefore,this paper proposes a study of image deformation methods for two-dimensional virtual try-on.In the virtual try-on deformation method,human pose estimation and clothing image segmentation are two important pre-processing steps.This paper establishes costume and human pose datasets,designs and implements human pose estimation and garment image segmentation,adopts a tree-structured structural model to pre-process pose estimation for human poses in datasets,adopts a color data model and minimizes iteration energy.The clothing image in the segmented data set.Experimental results show that the proposed method has higher accuracy in attitude estimation and segmentation.Considering the efficiency and cost of virtual try-on,a method of deformation of clothing image based on differential geometry energy is proposed.The purpose of aligning garments and users' images is achieved through the registration of garments;the garment is based on moving least squares.The deformation algorithm realizes the deformation of clothing according to the user's different postures,thereby providing the user with a trial effect under multiple postures.By solving the geometric energy function and optimizing the deformation energy,the costume deformation of the details can be truly realized,and the transformed clothing can be more realistic..Aiming at the problem that the virtual try-on effect is single and the garment details are notsufficiently dealt with,the MLS costume contour optimization method based on As-rigid-as-possible is proposed.The method uses the external control points of clothing and human posture profile as the Deformation constraints,further optimization and adjustment of the clothing contour,so that the clothing image can completely cover the human body to obtain better deformation results,and the deformed clothing image can still maintain the original clothing details as much as possible,to provide users with more Good virtual try-on experience.The proposed image deformation method for two-dimensional virtual try-on not only improves the user's virtual try-on effect under different body shapes and posture changes,but also ensures that the distorted clothing image maintains the original details of the clothing as much as possible,and is better applicable virtual try-on requirements in e-commerce.