Fast image component labeling on the GPU (Preliminary version)

Abstract—Image component labeling is a process that assigns unique labels to the connected components in a binary image. In this paper, we propose a fast image component labeling using a GPU. The key idea of our approach is to make a connected graph from connected components in sub-images and propagate these labels on that graph. We have implemented our algorithm on the NVIDIA GeForce GTX680. The experimental results for a spiral pattern image and a hilbert curve image show that our implementation is 1.7 and 2.9 times faster than that of the existing research, respectively. Index Terms—Image component labeling, GPU, CUD