Extraction Algorithm Based On Graph Theory Clothing Cut Image Foreground

Image segmentation, which is the foundation of image processing and analysis, has a wide range of algorithms, among which, the image segmentation algorithm based on Graph Cuts theory has become one of the hottest focus in research for recent year. With the rapid development of the network and the prevalence of online shopping, the web-based large-scale image retrieval system has attracted wider attentions. Image segmentation, as the most basic operation in image retrieval system, is of great importance because it plays a significant role in the effect of the image retrieval. As apparel online purchase takes a large part of online shopping, this paper mainly studies the apparel foreground extraction algorithm on the basis of the Graph Cuts theory, which is specified as follows:In this paper, the theoretical foundation and implementation framework of image segmentation based on Graph Cuts theory are studied at first, and the classic algorithms: Graph Cuts and GrabCut are implemented in the meantime. For the Graph Cuts algorithm, it’s essential for the user to sketch the foreground and background images using the brush, and then the images are segmented by the foreground/background probability model built in the light of the foreground/background pixel provided by the user. The algorithm lays a theoretical foundation for the image segmentation based on Graph Cuts theory. For the GrabCut algorithm, the gray histogram in the Graph Cuts algorithm is replaced with Gaussian mixture model to indicate color distribution, and the user only need to mark the initial "hard segmentation", which in return reduces the workload of human-computer interaction. Although the GrabCut algorithm works well on the image segmentation, it is neither suitable for the batch image processing, nor can it be used in apparel image retrieval system as each image requires manual labeling.Taking the above-mentioned disadvantages of the GrabCut algorithm into consideration, this paper proposes the following two solutions to realize the apparel image batch processing:generating the initial rectangle by fixing proportion and location and by region growing algorithm. By replacing manual labeling, the apparel images can be segmented automatically. Compared with many existing apparel extraction algorithm, this algorithm does not need any predefined model, or any manual labeling, which reduces the complexity of the work, and expands the range of applications. This paper analyzes and compares the three algorithms through experiments, and the final result shows that the GrabCut algorithm achieves the intended effect with a reduced workload and a better apparel foreground extraction effect.