A Region Based Segmentation using Pixel Block Fuzzy Similarity

In this paper, we propose a region based segmentation technique using a new similarity measure known as pixel block fuzzy similarity. Using this measure, an image is divided into pixel blocks and segmentation is performed by aggregating contiguous blocks having similar neighborhood. Our method is first used to interactively delineate a region via region growing. The results obtained are comparable and in some cases better than those from a classical region growing technique. Our scheme has also been adapted for automatic global image segmentation. Results show that our approach constantly provides fewer segments compared to watershed segmentation. Even though over-segmentation is observable at some edges, the proposed approach is generally suitable in delineating the main regions or objects in an image