Image Segmentation with Multiple Hypergraph Fusion and Superpixels
Kaixiang Wang · International Journal of Engineering Research and · 2018
This paper presents a new method of image segmentation based on superpixels and multiple hypergraph fusion.In this paper, the original image is oversegmented by the ultra-pixel segmentation algorithm SLIC, and the super-pixel blocks are extracted from the two aspects of color information and gradient information.A hypergraph model is constructed for each channel of the features.From the respect of random walk, the information of multiple hypergraphs is fused to construct the multi-hypergraph laplacian matrix.After obtaining the multi-hypergarph laplacian matrix, we construct the optimization model and solve the model by crossing the iterations.Then we can get the results of the superpixel blocks and the image segmentation results.With the superpixels and multiple hypergarph fusion, our method can reduce the loss of information effectively and make the segmentation more precise.Our method was evaluated in Berkeley Segmentation Database.The comparative experiments result with some state-of-the-art segmentation algorithms show that our method can give better segmentations for the natural images.