Image Segmentation Using SLIC Superpixels And Affinity Propagation Clustering

Image segmentation is a long-term research subject for computer vision, and it is very important for image processing and pattern recognition. The segmented regions have no overlapping areas, and the pixels in the same region share similar property. The object selection and parameter detection are benefited from image segmentation, which plays an important role in object recognition.In this paper, a new method of image segmentation is improved, named SLICAP, which combines the simple linear iterative clustering (SLIC) method with the affinity propagation (AP) clustering algorithm. First, the SLICAP technique uses the SLIC superpixels algorithm to form an over-segmentation of an image. Then, the color feature and texture feature of superpixels are merged in an asymmetry similarity matrix. In addition, the AP algorithm clusters these superpixels with the similarities obtained. Finally, space information of these superpixels is added by the trick which divides unlinked regions into different labels.Three similarity matrices are composed to find the most suitable one for SLICAP. Compared with the standard Ncuts method for image segmentation, the unsupervised improved SLICAP approach is relatively simple and fast, and there is no need to determine the number of targets. The experiments on the Berkeley segmentation database show that the image segmentation results produced by the improved SLICAP method are well consistent with the human visual perception. Quantitively, the improved SLICAP method outperforms other classical segmentation algorithms with the boundary-based and region-based criteria.An application of object recognition is designed for detecting the result of SLICAP approach, proving that it does well in object selection. The experiments of object recognition show that selecting object by SLICAP method could take the place of manual work in some ways.