基于Grab Cut和区域生长的服装图像前景提取算法

Foreground Extraction Algorithm Combined with Grab Cut and Region Growing for Clothing Image

针对Grab Cut算法应用于服装检索系统中服装图像的前景提取时所存在的交互式和复杂背景提取效果差问题,首先提出了一种全自动化的Grab Cut算法,以实现自动根据背景情况,对单一背景和复杂背景图像分别采用不同的方法生成初始矩形框并进行前景提取.由于该算法仍存在欠分割或过分割现象,故将该算法与区域生长算法相结合,给出了一种结合区域生长的全自动化Grab Cut算法.实验结果表明,无论对于单一背景还是复杂背景的图像,改进算法的前景提取效果都明显优于传统算法,不仅能准确获取服装前景区域,而且对于服装内部的过分割问题也有很大的改善.

Aiming at the problem of the interactive and complex background extraction effect of Grab Cut algorithm applied in the foreground retrieval of garment images in a garment retrieval system,a fully automated Grab Cut algorithm is proposed to adopt different methods to automatically generate the initial rectangle and then extract the foreground for single backgrounds and complex background images,depending on the background. As the algorithm still has under-segmentation and over-segmentation problems,a new fully automated Grab Cut algorithm combined with region growing algorithm is proposed. The results show that,the improved Grab Cut algorithm is superior to classical algorithms for both single background and complex background,and can not only accurately extract the foreground of clothing images,but also effectively reduce the over-segmentation within the images.