Abstract:
Image processing has always been an important and complex aspect of computer vision because of many factors. Some of the factors which makes image processing a complex process, includes images having different background and foreground colors, identifying object or sub objects within an image, similar object with different background and textures, distortion, scale, rotation, transformation, illumination or moving objects. The advancement in the field of computer technology, hand held devices and other technology led computer vision to a new world of horizon. Many theories and research work has been put forward in the field of computer vision. Computer vision field is separated into many different sub domains. Shape matching is an important subdomain for computer vision which relies on the shape representation and sophisticated feature space descriptors generation. Several approaches have been proposed for shape representation, matching, shape indexing and retrieval. Current approaches focus on generating sophisticated shape descriptors, not much attention has been given on the efficiency on shape matching and scalability of the system to handle larger dataset. Our proposed algorithm utilizes the feature space representation of shape, perform DFT on these features and construct a pruned Hierarchical indexing structure with no false negatives. For shape indexing and retrieval purpose, contour information is very important and how well we use the distance functions. In this thesis we present an approach using the boundary information for shape representation. The boundary information is invariant under rotation, distortion, noise and affine transformation. We utilize the contour data to generate a time series data for shapes. Shape matching is then carried out by using coefficient feature vector representation of time series data. We further enhance the search and retrieval process by constructing an indexed hierarchical tree structure for the shape dataset. To processes the indexing structure we use the multipath search technique which is robust under larger and scalable dataset. The experimental result shows that our proposed technique is time efficient, greater accuracy, robust and can be applied on larger dataset.
