Hierarchical Perceptual Segmentation using Arbitrary n-phase Levelsets

This paper presents a technique that is scaled, monotonic, and iterative distance mapping for levelsets implementation towards perceptual segmentation of images. It makes use of multiphase levelsets in which the speed of the evolving curve is controlled monotonically to improve the accuracy of segmentation, without affecting the speed of segmentation. The speed of curve evolution increases drastically through periodic reinitialization of levelset. Our method makes this reinitialization of the levelset functions affordable in controlling the evolution speed through a fast, efficient distance mapping technique, called Distance mapping using Scanning and Filling Technique (DSFT), without sacrificing the segmentation speed. This distance mapping technique, due to the fact that its time complexity is independent of the number of reference grid points, allows one to initialize the evolving curve arbitrarily, that further increases the segmentation speed. The image thus segmented through n-phase levelset segments perceptually through the perceptually uniform color space.