A Multiscale CCTA Plus Spectral Graph Partitioning for Image Segmentation

We proposed a multiscale image segmentation algorithm. In contrast to most multiscale image processing algorithms, this algorithm works on multiple scales of an image through connected coherence tree algorithm (CCTA) whose parameters can be changed to capture details in both coarse and fine level. By applying a graph-based technique, we can design a graph in which the nodes are both the regions and pixels produced by CCTA and the weights are the affinities between nodes. Finally, we run a spectral graph partitioning algorithm to partition on this graph to provide image segmentation. The experimental results on Berkeley image database demonstrate the accuracy of our algorithm as compared to existing popular methods.