Parallel Algorithm for Connected-Component Analysis Using CUDA

In this article, we introduce a parallel algorithm for connected-component analysis (CCA) on GPUs which drastically reduces the volume of data to transfer from GPU host. CCA algorithms targeting typically store extracted features in arrays large enough potentially hold maximum possible number objects given image size. Transferring these host requires portions overall execution time. Therefore, propose an uses CUDA kernel merge trees connected component feature structs. During tree merging, various properties, such as total area, centroid and bounding box, are accumulated. The structure then enables us only valid further processing or storing. Our benchmarks show that implementation significantly memory results whilst maintaining similar performance state-of-the-art algorithms.