Research On Interactive Image Segmentation

With the application of digital images widely used, a variety of image processing techniques, such as image segmentation, image enhancement, pattern recognition etc., play an increasingly important role in our lives. While image segmentation as the foundation and prerequisite of the other image processing techniques, plays a decisive role in image processing. Traditional image segmentation is usually an automatic process. However, due to the limitation of the processor’s comprehension, it is hard for automatic method to extract semantic objects from images. Therefore, this article will focus on the research of interactive image segmentation(IIS).Compared with traditional automatic method, the biggest advantage of the IIS method is that many difficult problems because of the ambiguity of data in the past become clearly defined by joining the subjective guidance from users. Finally, the problem can be solved accurately and efficiently by supplementing the targeted the algorithm. This article consists of three main points as follows:(1) The first point of this paper is the study of the IIS method based on statistical modeling and interactive graph cut segmentation. In order to facilitate the operation, drawing rectangle is used as the interactive manner to label the objects of interest. In this interaction model, the IIS problem can be solved effectively by combining statistical modeling for description of the global characteristics of the image and graph cut optimization method. Based on these ideas, this article presents two IIS methods: one is based on the Bayesian classifier and random walk method, and another is based on Bayesian classifier and shortest path algorithm.(2) In order to further improve the segmentation accuracy, this article also focus on multi-features fusion method for IIS algorithm. In this paper, we fuse the features of image, such as color, texture, gradient, and location, by using the method based on D-S evidence theory. Finally, in order to evaluate the proposed multi-feature fusion method, we combine the method with the proposed IIS algorithms to improve the segmentation accuracy.(3) In the third part, this paper focus on applying the proposed IIS algorithms to the mobile applications on the mobile platform. In order to adapt the limitations of software and hardware conditions of the mobile terminal, this article optimizes the proposed IIS methods by mainly addressing the two crucial issues, storage and computation. Finally we test the two IIS methods on mobile phones based on the Android OS about the real-time performance and practicality.All comparative experiments in this paper are conducted on the Berkeley dataset and the PIBS dataset. Experimental results show that compared with GrabCut and PIBS method, the proposed IIS methods have higher segmentation accuracy and efficiency. Furthermore, these two methods are nearly realtime and applicable for mobile phone applications.