Research On The Methods Of Virtual Try-On Based On Machine Learning

Online clothing sales have become a part of people's daily life,but consumers cannot try on clothes when they buy clothes online.Virtual Try-On effectively solves this problem by showing users the process of fitting.The simulation system constructed by Virtual Try-On truly and efficiently visualize the process of fitting to users.At present,Virtual Try-On not only needs to be improved in the reality of virtual fabric,but also in the efficiency of clothing comfort evaluation.In order to improve the reality and efficiency of the Virtual Try-On,this paper researches two directions of Virtual Try-On:cloning the real cloth into the virtual scene and improving the cloth comfort evaluation.In order to solve the above problems,this paper proposes a novel cloth material recognition algorithm using the enhanced cloth motion dense trajectory feature and a clothing comfort evaluation model learning from garment pattern database.1)Firstly,a material synthesis method is presented to construct the simulation video database with 64 kinds of cloth materials.Then,the feature information of each cloth material video is enhanced and the non-dynamic features are eliminated by transferring the pre-trained VGG network.Secondly,in order to capture and represent the dynamic features of the cloth simulation videos with different materials,the novel cloth motion dense trajectory feature is calculated.Finally,the feature database of cloth dynamic information can be created by coding their fisher vectors,and the SVM classifier can also be trained to set up the mapping of dynamic information of cloth motion video to material attribute parameters.2)Firstly,the sizes of mannequins and the graphs of garment patterns are collected,and the graphs of garment patterns are improved by using transfer learning to create garment pattern database.Secondly,we present a comfort label acquisition method based on Virtual Try-On,which adds comfort label to the corresponding graph of garment pattern.Thirdly,local binary pattern is extracted from the graph of garment pattern,and it is combined with the sizes of the corresponding mannequin to form clothing comfort feature vector.Finally,we extract the clothing comfort feature vectors of garment pattern database to train support vector machine.The main contributions of this paper are as follows:(1)A enhanced dynamic dense trajectory feature is proposed.The enhanced dynamic dense trajectory feature effectively represents the fabric motion feature by extracting fabric motion information.This feature effectively distinguishes the movement of different fabric materials,which is helpful for the identification of fabric material recognition model.(2)A fabric material recognition algorithm based on the enhanced dynamic dense trajectory feature is proposed.This algorithm effectively identifies the fabric material,and clones it into the virtual scene.The accuracy of this algorithm in identifying 64 kinds of material is 73.83%,which effectively clone the real fabric material.(3)A clothing comfort evaluation model learning from garment patterns based on transfer learning and support vector machine is proposed.This model effectively evaluates the clothing comfort and improves the clothing comfort evaluation of Virtual Try-On.The accuracy and average time of this algorithm are 83.4%and 12s respectively,which has high accuracy and efficiency.