Stopping criterion for anisotropic image diffusion

Abstract

Coherence-enhancing diffusion filtering is a striking application of the anisotropic diffusion in image processing. The technique deals with the problem of completion of interrupted lines and enhancement of flow-like features in fingerprint images. However, an anisotropic diffusion process is an iterated process, initializes with a poor quality image, and converges at the end towards a completely blurred image, with no structure surviving at the end. In anisotropic diffusion, one important question is how to find boundary between the under-smoothing and over-smoothing regions of the anisotropic diffusion process. The entropy change is found to be one such measure to describe that boundary adequately and thus provides a reasonable stopping rule for anisotropic diffusion. Numerical experiments with test pattern images confirm the desirable qualities of gap-closing and flow-enhancing qualities, along with the identification of frontier of useful smoothing. The proposed scheme is evaluated with the help of simulated images, and compared with other state of the art schemes using an objective criterion.