Fast image region growing

摘要

In this paper, we study the region growing approach to segmentation. We examine the effect of the merge criterion on region growing processing time, and propose a fast-merge policy that provides fast parallel segmentation. Fast-merge not only minimizes the number of merge rejections and results in fast region growing processes, but it also enhances the quality of segmentations. Our algorithm was tested on the BBN TC2000 shared memory multiprocessor system. It provides a well balanced load on the multiprocessors, and processing time is consistently fast on a wide range of images.