An Improved Fast Watershed Algorithm based on finding the Shortest Paths with Breadth First Search

Abstract

A watershed based on rainfall simulation is a proven technique for image segmentation. The only problem associated with it is the path regularization for pixels in the plateau. As the existing methods employ sequential techniques, the complexity of the algorithms remains high due to repetitive scanning of pixels. We propose an iterative method for finding the shortest and steepest path based on Breadth first search (BFS), which addresses the path regularization problem eliminating the repetitive scans. Experiments show, that the proposed algorithm significantly reduces the running time without compensating the performance when compared with the fastest known algorithm. General TermsFast watersheds, Image segmentation, Breadth First search, shortest path, path regularization. INTRODUCTIONImage segmentation is a fundamental problem in image analysis. The segmentation process can rely both on the uniformity of features within the regions or on edge evidence. Idea of watershed algorithm was developed from the field of topography. If we consider the analogy of a landscape and rain, water will find the swiftest descent path until it reaches some lake or sea, we can envisage lakes and seas correspond to regional minima [7]. The landscape can be completely partitioned into regions which draw water to a particular sea or lake. These regions are influence zones of the regional minima in an image which forms catchment basin. Though there are lot of variations and combinations with other methods, watershed still remain in research to improve either its time complexity, reducing the problem of over segmentation, and solving the path regularization. We, in this paper address all the three problems of watershed with our proposed algorithm.