SLIC Research and Implementation of a Parallel Optimization Algorithm

Abstract

SLIC is an image segmentation method that clusters picture pixels into superpixel blocks. But it has many flaws, including large amount of computation and high data dependency. In this paper, the reference point of the main computation of the SLIC algorithm is moved from the cluster center to the ordinary pixel point by changing the structure of the algorithm. Although the computation time increased, the paralism of the algorithm improved, and the pixel dependency reduced too. At the same time, the traversal method is changed from a single pixel point to a cluster center when calculating the pixel sum of each cluster center. The parallel optimization method is a hybrid programming paradigm based on OpenMP and MPI. Compared with the standard SLIC method, in this parallel algorithm, the computation time largely reduced in the multi-core computer.