Using Multi-layer Random Walker for Image Segmentation

Image segmentation can be considered as a label decision problem which assign different labels to every pixel according to its features. In this paper, we propose a supervised and interactive image segmentation algorithm. In our approach, we construct a new graph model which consists of a super-pixel layer and a high order layer. The super-pixel layer is composed by over-segmentation regions called superpixels and the high-order layer is generated by combining edge detection and these over segmentation regions. Then we construct a graph model and use a random walk algorithm to find the maximum probability label value for each superpixel. The proposed method shows very satisfactory results for sonic natural images.