Optimal partial shape similarity
Image and Vision Computing
| Hi-index | 0.00 |
Shape based image matching has been approached from different aspects. In our research, we developed end-to-end processes and algorithms that achieved very promising results for the MPEG-7 Shape Core Experiment. We have developed three algorithms that address curvature extraction, shape representation and shape similarity retrieval. Our Adaptive Contour Tracing Algorithm is able to trace open, closed and distorted shapes and output an ordered set of boundary points that can be used for any polygonal approximation method. Our multi-resolution curvature-based shape representation is efficient in terms of time and space complexity and flexible with minimum mathematical processing for fast real-time processing. Our 2-stage shape matching method is built over our shape representation method and consists of a data-driven search algorithm to exclude dissimilar shapes followed by a linear match algorithm that matches features and their neighbors. We have introduced a new fuzzy similarity measure that grades the similarity between shapes percentage wise and that is, unlike distance measures, closer to human understanding. Our shape representation is distinctive per object, similar for similar shapes, compact and efficient to compute, store and retrieve, and invariant to transformations such as rotation, uniform scaling and translation.