A Computer Vision Framework for Automated Shape Retrieval

With the increasing number of images generated every day, textual annotation of images for image mining becomes impractical and inefficient. Thus, computer vision based image retrieval has received considerable interest in recent years. One of the fundamental characteristics of any image representation of an object is its shape which plays a vital role to recognize the object at primitive level. Keeping thisview as the primary motivational focus, we propose a shape descriptive frameworkusing a multi-level tree structured representation called Hierarchical Convex Polygonal Decomposition (HCPD). Such a frameworkexplores different degrees of convexity of an object’s contour-segments in the course of its construction.The convex and non-convex segments of an object’s contour are discovered at every level of the HCPD-tree generation by repetitive convex-polygonal approximation of contour segments. We have also presented a novel shape-string-encoding schemefor representing the HCPD-tree which allows us touse the popular concept ofstring-edit distance to compute shape similarity score between two objects. The proposed framework when deployed for similar shape retrieval taskdemonstrates reasonably good performance in comparison withother popularshape-retrieval algorithms. Index Terms—Convex Polygon; Content BasedShape Retrieval; Shape Representation