Fast Nonparametric Image Segmentation with Dirichlet Processes

Among nonparametric clustering methods Dirichlet processes mixture models have proven to be very effective for unsupervised clustering. Image segmentation is an area where clustering has become a frequently used method. Many existing cluster type segmentation algorithms face problems such as slowness or parametric nature. We propose an effective method based on variational Dirichlet processes to achieve a great speed. In our approach we apply kd-tree to partition images and Dirichlet processes to cluster pixel color values in those partitions. Our experiments have shown that our method of clustering is fast compared to other methods of clustering using Dirichlet processes and also well performing compared spectral clustering.