A statistical hypothesis test-based image segmentation for low-bit rate coding

Abstract

We proposed a new image segmentation algorithm, called "SC-SAM", which checks the homogeneity of an image block using a statistical hypothesis test. SC-SAM consists of five processes: a split process, edge region adjustment, a merge process, postprocessing, and region representation. ShortCut test is applied to split a block as well as to merge two homogeneous regions into a region. A threshold value for the region homogeneity test can be chosen theoretically. SC-SAM can provide relatively very low computational complexity as well as keep the quality of a reconstructed image. Furthermore, SC-SAM removes the necessity of a control map used for refining the output in conventional algorithms. SC-SAM can considerably reduce the number of merged regions and computational time, while retaining the visual quality of the reconstructed image.