ROBUST SHAPE AXES EXTRACTION USING MULTIRESOLUTION GRADIENT VECTOR FIELDS

摘要

This paper describes a novel multiresolution technique for extracting the shape axes of 2D binary images. A gradient vector field representing the shape is generated using a multiresolution pyramidal technique and the shape axes are subsequently extracted by detecting the local directional disparity of the gradient vectors in the vector field. We discuss how the characteristics of the gradient vector field can be controlled such that the extracted shape axes in the interior of the shape are made robust to boundary noise and texture. Experimental results demonstrate this feature along with the scale-invariant nature of this technique. Its performance compares favorably to other commonly used skeletonization algorithms when applied to the problem of thinning binary images of typical textual symbols.