Research On Collaborative Segmentation Method For Clothing Image

With the rapid development of electronic commerce,the number of website commodities is increasing rapidly with explosive characteristics,and the number of clothing images on the web site is difficult to count.However,it has become a great challenge to accurately obtain the clothing images that you are interested in on the website.Garment image segmentation is the basic task of garment image retrieval technology.Due to the diversity and complex background of clothing images in practical applications and the fact that there are many situations in which multi-target garment and occlusion are involved,the accuracy of garment image segmentation is low.Therefore,a co-segmentation method for garment images is studied in this paper.In this paper,a co-segmentation algorithm for single object image based on regional active contour is proposed to solve the co-segmentation problem of the single-target common object in the image set.The specific steps of this algorithm are as follows: firstly,locate the potential are a of the garment through the upper body detection;secondly,the visual salient regions in each image is detected by transmitting the multiple upper body detection results;then,the upper body detection and the explicit detection are combined to roughly locate the clothe area in each image;finally,a Gaussian mixture model of the clothing area and background is modeled,and the common garment area in multiple images is extracted automatically.In view of the complex background and occlusion of multi-target clothing images,which results in a low accuracy of segmentation,a multi-target clothing image co-segmentation method is proposed in this paper.Firstly,the salient region features of the input multi-target clothing images are extracted in combination with an auxiliary data set,and the multi-target clothing image are segmented preliminarily.Then,construct a possible region map structure for preliminary segmentation results and the small image sets containing similar objects,and the images is co-segmented to obtain candidate multi-region clothing features.Finally,the output image segmentation results are output through the similarity calculation between multi-region clothing features and the common object segmentation.In order to solve the problem of complex and similar scene in the garment images which would the segmentation results,a co-segmentation algorithm for garment images in complex scenes is reposed in this paper.First,set the target clothing object region;then the contour curve of the target region is optimized to obtain the basic region contour of the target garment object;finally,based on the co-segmentation models in the third and fourth chapter,the single target clothing images and multi-target clothing images are co-segmented respectively.The method proposed in this paper realizes the optimization of the methods proposed in Chapter 3 and Chapter 4,and solves the problem that clothing images have similar backgrounds and edges are over-segmented.This proposed method has high segmentation accuracy and efficiency.It can be applied to the co-segmentation of single target and multi-target garment images in complex scenes,and can meet the needs of garment segmentation in simple scenes,single target,multi-target and similar scenes.