Error-Free Arithmetic and Architecture for H.264

Abstract

H.264/AVC is the newest international video coding standard. The standard uses a new DCT-like integer transform (Int-DCT) which is an integer orthogonal approximation to the classical DCT. The objective of using integer coefficients is to provide a completely reversible transform and also reduce the implementation complexity. The main disadvantage is that it increases the quantizer complexity and, consequently, the hardware costs. In this paper we present a new error-free encoding, based on algebraic integers, for a classical DCT and its hardware implementation. This can be used as an alternative transform for H.264 encoders, which results in a simple, parallel, and multiplication-free architecture, and also minimizes the complexity of the quantizer and the dynamic range of the computations. We propose an initial FPGA implementation where the system can process around 1650 Mpixel/sec, and satisfies all real time constraints for digital multimedia applications