Interactive Image segmentation by Dynamic Region Merging

In this paper dynamic region merging algorithm is proposed for interactive image segmentation. A part of image of user's interest is extracted from the input image and then segmentation is performed. This can be done by iteratively merging the regions according to some criteria. There are two issues in region merging algorithm: order of merging and stopping condition. In the proposed algorithm, the Sequential Probablity Ratio Test (SPRT) and minimal cost criterion are used to solve these two issues. The color image is converted to grayscale image and regions are merged if there is a proof for merging according to this predicate. The principle of dynamic programming is used to indicate the merging order. Experiments on different images are conducted to demonstrate the performance of the proposed dynamic region merging algorithm for interactive segmentation.