A Fast CUDA-Based Implementation for the Euclidean Distance Transform

Abstract

In Image Processing efficient algorithms are always pursued for applications that use the most advanced hardware architectures. Distance Transform is a classic operation for blurring effects, skeletonizing, segmentation and various other purposes. This article presents two implementations of the Euclidean Distance Transform using CUDA (Compute Unified Device Architecture) in GPU (Graphics Process Unit): of the Meijster's Sequential Algorithm and another is a very efficient algorithm of simple structure. Both using only shared memory. The results presented herein used images of various types and sizes to show a faster run time compared with the best-known implementations in CPU.