Abstract: In allusion to the traditional contour tracing arithmetic which can only work on binary images and is not able to handle the interlinking phenomena among objects, this paper introduces facula diffusion model based on the principle of human vision, and illustrates its controlling parameters including approximation, closing, length-limiting and hit-rate. This model works directly on grayscaled images and can automatically separate the interlinked objects. Taking advantage of the facula diffusion model, three new feature indices including acreage, abnormity and fluctuation are composed to describe the object shape in its dimension, the departure from roundness and the roughness of its contour respectively. It turns out that the new model can extract the contour of fiber cross-section accurately and the feature indices built on this model can reflect the shape characters of the object in an effective way.