Research And Application Of Online Selection Based Image Segmentation

With the development of Computer Vision in the past few years, images become more and more important both in our daily life and in the work. So, techniques of segmenting the interested region from an image play a crucial role especially when understanding and analyzing the content of images.This paper proposes a new framework for image segmentation, which select feature/parameter adaptively. Our work is mainly on three parts: (1) this paper implements the work of modeling background/foreground with GMM based on traditional max-flow/min-cut algorithm, thus overcomes the problem of single source/sink. (2) This paper introduces the online feature selection to graph cut, which adaptively gives a feature image with best dissimilarity between foreground and background. This operation gives better segmentation result. (3) This paper proposes a new strategy for the parameter selection, by searching and judging the results derived from different parameters on a small parameter space.This paper evaluates its method on IG02 dataset, which gives better precision and recall rate, compared with traditional segmentation method. At the end of this paper, the proposed method is used to a new field of precise forensic detection, which also gives pleasing results.