摘要
为了实现服装图像自动分割处理,本文提出基于Faster R-CNN模型结合GrabCut的图像分割方法。利用快速区域卷积神经网络的基本框架,将街拍图的待检测任务分为上衣、裙子、包等六个类别,对原有的基本框架模型的全连接层参数进行调整,得到前景目标框作为GrabCut分割算法的初始框,再使用GrabCut算法进行服装区域提取,从复杂背景的图片中定位服装位置,去除复杂背景,实现服装区域分割。实验结果显示,本文方法能够很好的实现服装自然轮廓检测和提取,适用于图像局部弱轮廓边缘的检测及大批量服装图像分割处理,并且可供大批量处理图片时选择性自动款式类别提取,提高了服装图像分割处理的效率。
In order to realize the automatic segmentation of clothing images,an image segmentation method based on Faster RCNN combined with GrabCut is proposed.Firstly,the basic framework of the Fast R-CNN is used to subdivide the to-be-detected tasks of street photographs into six categories:tops,skirts,bags,etc.Next,after adjusting the model full connection layer parameters based on the original basic framework,the foreground object box is obtained as the initial frame of the GrabCut segmentation algorithm.Then we use GrabCut algorithm to extract garment area.The method locates the clothing position from the picture of the complex background,removes the complex background,and realizes the segmentation of the clothing area.The experimental results showthat the proposed method has good natural contour detection and extraction ability,and is suitable for the detection of local weak contour edges of images and the processing of large-scale clothing image segmentation.Beyond that,it can be used for selective automatic style category extraction in large batch processing of images.It improves the efficiency of the garment image segmentation process.
作者
杨思
徐增波
陈冲
YANG Si;XU Zengbo;CHEN Chong(Shanghai University Of Engineering Science Costume Institute,Shanghai 201620,China)
出处
《智能计算机与应用》
2020年第7期306-310,共5页
Intelligent Computer and Applications
基金
上海市科学技术委员会科技创新行动计划资助项目(18030501400)
研究生创新项目(E3-0903-18-01185)
