An Optimized Union-Find Algorithm for Connected Components Labeling Using GPUs

In this paper, we report on an optimized union-find (UF) algorithm that can label the connected components on a 2D image efficiently by employing GPU architecture. The proposed method comprises three phases: UF-based local merge, boundary analysis, and link. The coarse labeling in local merge, which makes computation efficient because the length of the labelequivalence list is sharply suppressed . Boundary analysis only manages the cells on the boundary of each thread block to launch fewer CUDA threads. We compared our method with the label equivalence algorithm [1], conventional parallel UF algorithm [2], and line-based UF algorithm [3]. Evaluation results show that the proposed algorithm speeds up the average running time by around 5x, 3x, and 1.3x, respectively.