Research On Clothed-body Segmentation Technology Based On Deep Learning

Clothing image segmentation technology is widely used in clothing retrieval,clothing recommendation,virtual fitting and other fields.More and more researchers use semantic segmentation network based on deep learning to segment clothing images,and gradually become a hot topic in the field of computer vision.Due to the various attributes of clothing,different fabric styles and textures,and the human posture and complex scene in the clothing image are easy to block the clothing.After the clothing image is segmented by most semantic segmentation networks based on deep learning,there will be some problems,such as rough clothing edge segmentation,poor segmentation accuracy,clothing occlusion and insufficient extraction of deep semantic features of clothing.Based on the existing semantic segmentation network based on deep learning,this paper has carried out the following work on how to improve the segmentation effect of clothing image:(1)Aiming at the problem of rough clothing edge segmentation in clothing image segmentation based on deep learning semantic segmentation network,the coordinated attention mechanism is introduced into the backbone network of Deeplab v3+ network,and the coordinated attention mechanism is used to capture the location information and channel relationship,so as to capture the characteristics of the learning target area,which is conducive to improve the segmentation ability of the network to the fine parts of clothing image.(2)Aiming at the problems of poor segmentation accuracy and insufficient extraction of deep semantic features of clothing image segmentation in the semantic segmentation network based on deep learning,the semantic feature enhancement module is introduced into Deeplab v3+ network,and the semantic feature information of clothing is enhanced by applying the nonlocal attention block to different size feature maps,which improves the accuracy of clothing segmentation to a certain extent.(3)Aiming at the problem of blocking clothing after deep learning semantic segmentation network segmenting clothing,the PatchMatch algorithm is used to deal with the occlusion area of clothing,which can deal with the problems such as arms blocking clothing,and achieve good visual effect.(4)A clothing image segmentation system is designed and implemented.The system uses the semantic segmentation network trained on the clothing data set to segment the clothing image,uses the PatchMatch algorithm to remove the occlusion of the clothing image,and finally obtains the target clothing image.This paper improves the existing semantic segmentation network based on deep learning.The experiments show that the mIoU and MPA of Deeplab v3+ network embedded with coordinated attention mechanism and semantic feature enhancement module are improved by2.1% and 2.3% compared with Deeplab v3+ network on deepfashion2 dataset.Finally,this paper designs and develops a clothing image segmentation system,which not only integrates the trained convergent clothing segmentation network to segment the clothes,but also adds the PatchMatch algorithm to remove the occlusion of the clothing and extract a purer clothing image.