A Multi-level Polygonal Approximation-Based Shape Encoding Framework for Automated Shape Retrieval

Abstract

Due to exponential growth of image data, textual annotation of every image object has failed to provide practical and efficient solution to the problem of image mining. Computer vision-based methods for image retrieval have been drawing significant attention with the progress in achieving fast computational execution power. One of the fundamental characteristics of an image object is its shape which plays a vital role to recognize the object at a primitive level. We have studied the scope of a shape descriptive framework based on multi-level polygonal approximations for generating features of the shape. Such a framework explores the contour of an object at multiple approximation stages and captures shape features of varying significance at each approximation stage. The proposed algorithm determines polygonal approximations of a shape starting from coarse-level to more refined-level representation by varying number of polygon sides. We have presented a shape encoding scheme based on multi-level polygonal approximation which allows us to use the popular distance metrics to compute shape dissimilarity score between two objects. The proposed framework when deployed for similar shape-retrieval task demonstrates fairly good performance in comparison with other popular shape-retrieval algorithms.