A Line-Based Connected Component Labeling Algorithm Using GPUs

Connected Component Labeling (CCL) is a type of basic images analysis, and it is used by many applications. This paper proposes a new parallel algorithm of CCL with GPU architecture for the binary image. We improved the Union-Find technique of CCL using GPGPU and developed the parallel algorithm to run fast in divided areas of each one line. We compared proposed algorithm with Union-Find technique and the Label Equivalence technique by an experiment using the CUDA. Our approach was a 1.61x speedup compared with Union-find when runs on the large image.