Research On Fast Color Image Segmentation Method Based On Superpixel

Image segmentation is a fundamental and important step in the field of computer vision with the purpose of separating and extracting the region of interest(ROI)from other parts of the image.Superpixel is a topic that has been widely concerned by researchers in recent years.Compared with the traditional pixels,superpixel can effectively reduce the computational complexity of the segmentation method,and improve the final segmentation results.In this dissertation,superpixels is used for image preprocessing,and a fast color image segmentation method based on superpixel is proposed.The main work of this dissertation is as follows:1.A novel superpixel segmentation method based on edge probability is proposed.Due to the serious over-segmentation of traditional superpixel segmentation method,it is still very necessary to further optimize and purify the existing superpixels in the image preprocessing stage.Therefore,the original image is detected via an excellent structured edge detection operator.And then,through an image binarization method which is based on global mean and global standard deviation,the threshold operation is performed to get the edge binary image for purifying the over-segmentation superpixel generated by the traditional superpixel segmentation method.Experimental results show that the proposed method can effectively purify superpixels and reduce the number of initial superpixels,and the purified superpixels possess good boundary consistency.2.An interactive image segmentation method based on superpixel merging is proposed.The efficiency of superpixel merging method depends on the initial number of superpixels to some extent,therefore,the superpixels can be further refined and the number of iterations during the superpixel merging process can be effectively reduce so that the computational efficiency of the algorithm is improved.Meanwhile,the information of color and texture is used to measure the similarity between superpixels.Finally,a fast and efficient superpixel merging is realized by making use of region adjacency graph and nearest neighbor graph.Simulation experiments show that the proposed method is better in segmentation performance and valuable in practical applications.3.An unsupervised image segmentation method based on superpixel and spectral clustering is proposed.Because the traditional spectral clustering algorithm has high computational complexity during feature decomposition,and the number of superpixels in the image is far less than the number of pixels,therefore the efficiency of segmentation can be improved by using superpixels on spectral clustering.Furthermore,to address the problem that spectral clustering requires the number of clusters manually,a novel spectral clustering algorithm based on density peak optimization is proposed,which can determine the number of clusters and the initial cluster centers adaptively.Experimental results show our method is more competitive to other segmentation algorithms.