Parallel Method of Signed Distance Function Based on Normal Radiation

Abstract

Signed distance function has important application in image processing such as level set based image segmentation and feature extraction. With higher and higher image resolution, the computational efficiency of signed distance function directly effects the computational time of image processing. To reduce the time cost, an improved source scanning method was proposed. This method first calculated the signed distance to the close curve of the pixels on the normal line, then calculated the signed distance of all the pixels based on source scanning method, which can reduce the iteration time and improve the computational efficiency. This method was also performed in parallel way on CUDA. Experiments show the proposed parallel method is faster than lower envelope based method and the Voronoi diagram based method.