FRAME BIT ALLOCATION FOR H.264 USING CAUCHY-DISTRIBUTION BASED SOURCE MODELLING Nejat Kamaci and Yucel Altunbasak Center for Signal and Image Processing Georgia Institute of Technology Atlanta, GA 30332 email:{kamaci,yucel}@ece.gatech.edu ABSTRACT Based on the observation that a Cauchy density is more accurate in estimating the distribution of the AC coefficients than the tra- ditional Laplacian density for H.264 video coders, a rate model with improved accuracy is derived. Using the new rate model, a frame bit-allocation algorithm for the rate control of an H.264 video coder is developed. Extensive analysis with carefully se- lected anchor video sequences demonstrates a 0.43 dB average PSNR improvement over the JM 8.4 rate control algorithm, and a 0.27 dB average PSNR improvement over the TM5-based bit- allocation algorithm that has recently been proposed for H.264 by Ma et al. The analysis also demonstrates 20% and 60% reductions in PSNR variation among the encoded pictures when compared to the JM 8.4 rate control algorithm and the TM5-based bit-allocation algorithm, respectively. 1. INTRODUCTION Over the past few decades, transform-based compression for im- age and video sources has gained widespread popularity for visual information management, processing, and communications. As a result, several industry standards have been developed, such as MPEG-2 [1] and H.264 [2] for video coding. With all of these image and video processing methods, the image frame(s) is di- vided into nonoverlapping blocks, and a transformation is applied to the block before quantization and entropy coding. The two- dimensional discrete cosine transform (DCT) is the most common transform used in these methods 1 . The H.264/AVC (equivalently MPEG-4 Part 10) standard is one of the most advanced video coding standards that has been de- veloped. H.264/AVC features a number of new technologies such as intra prediction, integer transform and multi-frame prediction, in addition to improvements on the existing technologies. In com- mon with the earlier standards, the H.264/AVC does not explicitly define an encoder-decoder pair. Many functional parts of the en- coder and the decoder are left open for optimization. One of these functional parts is the rate-control module that is responsible for controlling the output bit rate of the encoder. 1.1. Rate control for H.264/AVC The output bit rate and video quality of a video encoder depend on several coding parameters such as the quantization scale (Q) and 1 The H.264 coder uses an integer transform that is a close approxima- tion to the DCT. coding mode. In particular, choosing a large quantization scale re- duces the resulting bit rate, while at the same time reducing the visual quality of the encoded video. In most applications, a pre- determined constant output bit rate is desired. These applications are referred to as constant bit-rate (CBR) applications. The output bit rate of an encoder can be controlled by carefully selecting the quantization parameters for each coding block. This task is per- formed by the rate-control module. The goal of rate-control unit is to keep the output bit rate within constrained limits while achiev- ing maximally uniform video quality. For practical reasons, the rate-control problem is usually stud- ied in three subproblems: (i) GOP bit allocation, (ii) picture bit allocation, and (iii) macroblock Q selection. GOP bit allocation involves selecting the number of bits to allocate to a GOP, which in the case of CBR rate-control, simply amounts to assigning a fixed number of bits per GOP. Picture bit allocation involves distributing the GOP budget among the picture frames, so as to achieve a max- imal, uniform video quality. Macroblock Q selection involves tun- ing the Q parameter for each macroblock of a frame so that the the rate regulations are met and a uniform quality is achieved within the picture. As in the H.264 reference software, Q selection may also affect the motion estimation and compensation operations. Many conventional rate-control algorithms use rate and distor- tion models for their operation. The performance of a rate-control algorithm greatly depends on its ability to estimate the rate and the distortion. In our earlier study we have demonstrated that the accu- racy of modelling rate-distortion relation can be improved by us- ing a Cauchy-distribution fit to the DCT coefficients, especially for the H.264 video coders [3]. The Cauchy distribution’s heavy tails help estimating the actual statistical distribution of the transform coefficients more accurately for larger values. Consecutively, it helps improving the accuracy of the rate and distortion models. In this study, we demonstrate the benefit of using Cauchy-distribution based rate model in frame bit allocation for rate control purposes. 2. RATE MODELLING USING CAUCHY-DISTRIBUTION The entropy of a uniformly quantized Cauchy source with param- eter µ and quantization level Q is given in [3] by H (µ, Q) = − 2 π ξ  1 2 , 2µ, Q  log 2 ξ  1 2 , 2µ, Q  − 2 π ∞  i=1 ξ (i, µ, Q) log 2 ξ (i, µ, Q) , (1) II - 57 0-7803-8874-7/05/$20.00 ©2005 IEEE ICASSP 2005