Image Segmentation Based On Saliency Detection And Diffusion Model

With the development of multimedia technology and digital imaging device, image analysis and understanding has attracted the attention of many researchers. As a basic and one of the key steps in image analysis, image segmentation has become the focus and hotspot in the research of the image engineering technology. Image segmentation is aimed to partition image into different areas with similar characteristics by extracting the feature information of each area. As a more challenging topic, image co-segmentation is to simultaneously divide a set of images into multiple separate meaningful regions which correspond to different object classes by mining the coherence in the image set. Inspired by the recent researches in the image segmentation field, this paper proposed an automatic image segmentation model which combines several features properly. Furthermore, this paper proposed a new image co-segmentation model by building the diffusion network between images which can establish connections with similar object in image co-segmentation. The proposed methods are compared with state of the art methods on benchmark dataset. The contributions of this paper can be concluded as follows:Firstly, this paper introduced the basic theories of image saliency detection which are used by subsequent proposed image segmentation method. Harris salient point detection and Region Contrast(RC) saliency detection method were discussed in this paper. For further comparison with Region Contrast(RC) saliency detection method, 9 state of the art saliency detection methods are compared on the MSRA-1000 benchmark dataset.Secondly, this paper proposed an automatic shape prior image segmentation method which integrates salient region with affinity propagation clustering(SRAPC). The proposed method adopted the saliency region and Harris salient point method to predict the salient points on the foreground. With AP clustering, the shape stars were selected and the shape prior energy function was solved under the framework of Graph Cuts. The proposed automatic segmentation method is evaluated on benchmark datasets and experimental results support the proposal.Furthermore, it is difficult to segment the common objects from complicated background in image co-segmentation because there were always weak connections built between objects. Addressed this problem, we established a 3D diffusion network between images by connecting conduction edge with similar objects. Then, image co-segmentation is converting into how to get the most marginal gain in the conducting network. It is proved that the problem could be solved by sub-modular theory. Compared with several state of the art co-segmentation methods, the experimental results of the proposed method show good performance on benchmark datasets.