A fast region merging algorithm for watershed segmentation

This paper presents a new fast method for region merging after watershed segmentation. We name it watershed collapsing algorithm (WCA) as its merging operation mimics the watershed collapsing and water flooding process to a great extent. The three stages of the algorithm are discussed in detail: initialization, region dissimilarity estimation, watershed collapsing and region flooding. In region dissimilarity estimation stage, Butterworth low-pass filter is adopted to improve the region dissimilarity function, which estimates the similarity of adjacent regions. The kernel of whole algorithm in WCA is the merging method based on a flooding process analogy. Experimental results on various images demonstrate the efficiency of the presented algorithm and statistics of execution time indicate that the proposed algorithm is much faster than region-based methods.