Systolic array based motion estimation architecture of 3D DWT sub band component for video processing

Abstract

This paper describes Full Search Block Matching Algorithm (FSBMA) performed on 3D image in the transformed domain. An image of size N×N×8 is first transformed in to eight sub bands each of size N/2 × N/2 × 4 employing 3D Discrete Wavelet Transform (3D DWT). FSBMA performed on LLL sub band component which achieves good Peak Signal to Noise Ratio (PSNR), Compression Ratio (CR) and reduces computation complexity when compared with the original image. Modified lifting scheme based DWT and Systolic Array Architecture (SAA) based FSBMA are implemented on Field Programmable Gate Array (FPGA). 3D DWT implemented on FPGA operates at 230 MHz and consumes a power of 0.29 W, while FSBMA implemented on FPGA to reduce the computational time for motion estimation, operates at 120 MHz and consumes a power less than 0.34 W.