Shape Representation and Retrieval Using Distance Histograms

Among all the issues related to Content Based Image Retrieval systems, retrieving images based on their shapes is an important one. Many approaches exist utilizing shape representation and comparison, e.g., the methods based on Fourier descriptors. In this thesis, we propose a novel method for shape representation. In our method, we calculate the centroid of a shape and choose a set of sample points around this shape’s boundary. From those, we obtain a set of radii. We then use these radii lengths to construct a Distance Histogram as the representation of shape. The natural characteristic of our method makes itself invariant to rotation and translation. Furthermore, it can be made invariant to scale by a simple normalization. To evaluate our approach, we perform a large set of experiments on a database of shapes. Using either a database of synthetic shapes or a database of real shapes, we compare our method to that based on Fourier descriptors, which is a well-known and effective approach. The results of the experiments show that our method is an effective, economical, and flexible approach for shape representation.