الفهرس | Only 14 pages are availabe for public view |
Abstract In this thesis, a novel circle views shape signature for recognizing 2-D object silhouettes is proposed. Many views for the shape contour are defined from one circular orbit (or more) centered at shape centroid. One compact and robust descriptor is obtained by applying Fourier transform. The obtained descriptor is translation, rotation, and scale invariant. The proposed signature provides a promising retrieval rate (83.71% on MPEG-7 database). An attractive feature of the proposed signature is its simplicity and computational efficiency. A further increase in retrieval rate (90.35%) is achieved using a shape context learning technique. This thesis also proposes a hardware accelerator for feature extraction based on the proposed signature. The proposed hardware accelerator is implemented on FPGA by using Verilog HDL. A parallel architecture for the hardware accelerator is proposed. Using 16 processing elements, the proposed hardware achieved 85.35 times speedup (Artix-7 XC7A200T FPGA chip) compared to a software implementation. |