The H.264 Advanced Video Compression Standard, Second Edition

Chapter 7. H.264 transform and coding

Introduction

Chapter 6 described the 'front end' of an H.264 encoder, the prediction processes that remove some redundancy by creating and subtracting an estimate of the current block of image data. This 'front end' prediction stage is lossless, i.e. it is a process that is fully reversible without loss of data. However, H.264 is fundamentally a lossy compression format, in which a degree of visual distortion is introduced into the video signal as a trade-off for higher compression performance. This distortion occurs in the transform/quantization process. In earlier standards there was an obvious boundary between the transform, converting a block of image samples into a different domain, and quantization, reducing the precision of transform coefficients. As will become clear, this boundary is less obvious in an H.264 codec, with an overlap of the transform and quantization stages. This, together with the new approach of exactly specifying a reversible integer transform 'core', makes the H.264 transform and quantization stage significantly different from earlier compression standards.

After prediction, transform and quantization, the video signal is represented as a series of quantized transform coefficients together with prediction parameters. These values must be coded into a bitstream that can be efficiently transmitted or stored and can be decoded to reconstruct the video signal. H.264/AVC provides several different mechanisms for converting ...