基于卷积姿态机和GrabCut的服装图像轮廓分割方法  

Garment image contour segmentation method based on convolution pose machine and GrabCut

作　　者：游小荣[1] 李淑芳[1,2] 熊宗志 YOU Xiaorong;LI Shufang;XIONG Zongzhi(Changzhou Textile Garment Institute,Changzhou,Jiangsu 213164,China;Changzhou Key Laboratory of Eco-Textile Technology,Changzhou,Jiangsu 213164,China)

机构地区：[1]常州纺织服装职业技术学院,江苏常州213164 [2]常州市生态纺织技术重点实验室,江苏常州213164

出　　处：《毛纺科技》2022年第5期86-90,共5页Wool Textile Journal

基　　金：江苏省高等学校大学生实践创新训练项目(202112807027Y)。

摘　　要：GrabCut在服装图像分割中得到广泛应用,为解决GrabCut需要人工干预以及复杂背景下精度不高的问题,提出了一种基于卷积姿态机(Convolutional Pose Machines)和GrabCut的服装图像分割方法。借助卷积姿态机,使用包含服装的图片数据进行训练,得到卷积姿态机网络,从复杂背景中定位出服装关键点,形成包含服装的矩形框,并以此作为GrabCut算法的初始矩形框,经过GrabCut算法处理,从复杂背景分割出服装图像。结果证明:与传统需要人工干预的GrabCut方法相比,本方法无需人工干预,便可将复杂背景中的服装图像自动分割出来,在不损失精确度的基础上,可实现大批量服装图像的自动分割。Grabcut was widely used in garment image segmentation.In order to solve the shortcomings of manual intervention and low accuracy in complex background,a garment image segmentation method based on convolution pose machine and GrabCut was proposed.With the help of convolution pose machines,the image data containing clothing was used for training to obtain the convolution pose machines network.The key points of clothing were located from the complex background to form a rectangular box containing clothing,which was used as the initial rectangular box of GrabCut algorithm.After being processed by GrabCut algorithm,the clothing image was segmented from the complex background.Experiments show that this method can automatically segment clothing images in complex background without manual intervention.Compared with the traditional GrabCut method,this method can realize the automatic segmentation of a large number of clothing images without losing accuracy.

关 键 词：服装轮廓 图像分割 卷积姿态机 GRABCUT 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TS941.26[自动化与计算机技术—计算机科学与技术]