Robust Shape Retrieval through a Novel Statistical Descriptor

We propose a novel statistical descriptor, Multiple References Histogram Matrix (MRHM), for robust shape retrieval, especially for degraded shape images. For each shape image, MRHM first generates uniform grids and filters noises in each grid by line Hough transformations and curve-fitting transformations. Then MRHM selects a reference for each grid and calculates its local distribution between the reference point and the shape pixels. Finally, all the local distributions are integrated into a global distribution matrix for matching symbols. Experimental results on the MPEG-7 Shape Silhouette Database and the GREC2005 Shape Database show that the proposed method’s recognition rate for degraded shape images is greatly improved over a recent method (SFHM).