Incremental indexing and retrieval mechanism for scalable and robust shape matching

摘要

Techniques for efficient and effective content-based image matching are becoming increasingly important with the widespread increase in digital image capturing systems. Shape of an object, represented by its contour, is one of the most important visual feature that is thought to be used by humans to determine the similarity of objects. The selected feature and its distance measure must be robust to different distortions such as noise, articulation, scale and rotation. Existing approaches provides invariance to these distortions at the cost of either the accuracy due to poor discrimination ability or the efficiency. In this paper, we present an effective representation of shape, using its boundary information, that is robust to arbitrary distortions and affine transformation. The contour representation of shape is converted into time series and is modeled using orthogonal basis function representations. Shape matching is then carried out in the chosen coefficient feature space resulting in efficient matching. The efficiency of shape matching is further improved by indexing the shape descriptors using hierarchical indexing structure. A novel distributed beam search based technique is proposed that operates on the indexing structure and ensures no false dismissal for a given k-NN query. Experimental evaluation demonstrates that the proposed shape representation and matching mechanism is robust, efficient and scalable to very large shape datasets.