Low energy video processing and compression hardware designs

Official URL: http://risc01.sabanciuniv.edu/record=b1818140 (Table of Contents)

Digital video processing and compression algorithms are used in many commercial products such as mobile devices, unmanned aerial vehicles, and autonomous cars. Increasing resolution of videos used in these commercial products increased computational complexities of digital video processing and compression algorithms. Therefore, it is necessary to reduce computational complexities of digital video processing and compression algorithms, and energy consumptions of digital video processing and compression hardware without reducing visual quality. In this thesis, we propose a novel adaptive 2D digital image processing algorithm for 2D median filter, Gaussian blur and image sharpening. We designed low energy 2D median filter, Gaussian blur and image sharpening hardware using the proposed algorithm. We propose approximate HEVC intra prediction and HEVC fractional interpolation algorithms. We designed low energy approximate HEVC intra prediction and HEVC fractional interpolation hardware. We also propose several HEVC fractional interpolation hardware architectures. We propose novel computational complexity and energy reduction techniques for HEVC DCT and inverse DCT/DST. We designed high performance and low energy hardware for HEVC DCT and inverse DCT/DST including the proposed techniques. VII We quantified computation reductions achieved and video quality loss caused by the proposed algorithms and techniques. We implemented the proposed hardware architectures in Verilog HDL. We mapped the Verilog RTL codes to Xilinx Virtex 6 and Xilinx ZYNQ FPGAs, and estimated their power consumptions using Xilinx XPower Analyzer tool. The proposed algorithms and techniques significantly reduced the power and energy consumptions of these FPGA implementations in some cases with no PSNR loss and in some cases with very small PSNR loss.