Automatic foreground extraction of clothing images based on GrabCut in massive images

In recent years, clothing image retrieval has become an important research focus in the field of CBIR (content based image retrieval) [1]. Because of the complexity of CBIR, there are still many difficulties to be overcome. When people search a clothes, they usually focus on the clothing area. Therefore, we must remove unrelated background, or it will affect feature extraction results. Usually, foreground extraction is more time-consuming than extracting images' features. To establish a database of several million clothing images, it is very necessary to reduce time of extraction. In this paper, we proposed a fast method for extracting the clothing area automatically based on GrabCut algorithm [2]. Compared to extracting clothing area in image manually, auto extraction will significantly reduce workload. Firstly, we use a rectangle proportional to size of image instead of user input. Secondly, to solve the problem of time consuming, we did some optimization work. Experiment results show that an overall foreground extraction rate of 82.2% can be achieved without human interaction.