A new seeded region growing algorithm for large object segmentation
Yanhui Shan, Kun-Yu Tsai, Jian Wu · 2011
The original average contrast and peripheral contrast (ACPC) seeded region growing method performs well in segmenting small object, however, it is not effective to segment large objects with large intensity variations. In this paper, we improve the original ACPC method and propose a new seeded region growing method which is more adaptive to large objects segmentation. Every time, only one pixel which has the largest similar intensity with the average intensity of pixels in current region is picked as the new seed, and it will be added into current region. The optimal segmentation result is the one with the largest average contrast (AC) and peripheral contrast (PC). In the case of objects with serious intensity variation, we can initially set seeds both in high and low intensity regions to improve the segmentation result. In addition, we accelerate the algorithm to reduce the computation cost. The experiment results show that our method is good at segmenting large objects. Besides, the efficiency of the algorithm is greatly increased.