Joint optimization coding for level and map information in H.264/AVC

https://doi.org/10.1016/j.image.2011.07.002Get rights and content

Abstract

Since the joint optimization coding for the level and the map information of transformed coefficients has been applied to the image coding successfully, this paper addresses the problem of how to optimize the two parts of information of residual coefficients jointly in the state-of-the-art video coding standard—H.264/AVC. In order to solve this problem, this paper presents a pruning method to zero out residual coefficients, which are not valuable for encoding in rate-distortion optimization (RDO) sense. Furthermore, a joint optimization scheme is presented to combine scalar quantization with this pruning method together to further improve the compression performance of the proposed algorithm. Experiments have been conducted based on the reference encoder JM12.4 of H.264/AVC. Comparative studies show that the proposed joint coding method can achieve a PSNR gain more than 0.5 dB at a certain bit rate, compared with the coding method in the H.264/AVC. Especially, the PSNR gain can reach up to 0.9 dB, or equivalently, 17% bit rate saving can be achieved at some special cases.

Highlights

► Level information and map information of residual coefficients are optimized jointly. ► Pruning method is presented to zero out residual coefficients, which are not valuable. ► Scalar quantization and pruning method are combined together in the proposed scheme. ► Fast coding scheme is presented to reduce the complexity of the proposed scheme. ► Proposed scheme can provide better RD performance at high bit rate than H.264/AVC.

Introduction

In the image/video coding scheme, after transform and quantization, one nonzero coefficient to be encoded usually contains two parts of information: the level information, which is the sign and magnitude of this coefficient, and the map information, which is the position of this coefficient. How to encode the level and the map information efficiently is very important to the compression performance of the image/video coding scheme. The common method is to encode the two parts of information separately such as the entropy coding methods for residual coefficients in H264/AVC [1]. In this kind of method, more bits need to be assigned to encode the map information of nonzero coefficients. In order to solve this problem, many algorithms have been presented to encode the level and the map information of nonzero coefficients in a joint optimal way. Using these algorithms, the bits used for encoding the map information can be reduced and thus more bits can be assigned to encode the level information at a given bit rate. These algorithms can be classified into two categories up to now. One category is based on the wavelet transform in image coding. The typical works of this category are the presentation and the consummation of the space-frequency quantization scheme [2], [3], [4], [5]. This scheme combines frequency quantization (usually scalar quantization in frequency domain) with spatial quantization together to jointly optimize the coding of the level and the map information of wavelet transform coefficients. The other category is based on the matching pursuit (MP) image coding. The typical work of this category is the algorithm proposed recently in [6]. In MP image coding, the MP coefficients and dictionary indexes are usually encoded in position order or in magnitude order. The former method can reduce the bits used for the position information of dictionary indexes, while the latter method can result in fewer bits required to encode the MP coefficients. The algorithm in [6] presents a tradeoff between the two methods and takes advantage of the benefits of both encoding methods.

The idea of encoding the level and the map information in a joint optimal way has been applied to image coding successfully; however, it has not been applied to video coding standard yet. H.264/AVC, which is the newest video compression standard, has shown better compression performance than its precedents [7]. Many efforts are still being made to further improve the rate-distortion (RD) performance in the H.264 compliant coding environment [8], [9], [10]. In H.264/AVC, the residual coefficients that have been transformed and quantized are directly entropy encoded. This method does not take the joint optimization coding for the level and the map information of residual coefficients into consideration. This paper addresses the problem of how to jointly optimize the two parts of information in H.264/AVC.

The remainder of this paper is organized as follows. In Section 2, the coding process of residual coefficients in H.264/AVC is introduced first. Then the defect existing in this coding process will be discussed. Section 3 describes the proposed pruning algorithm and the joint optimization scheme. Section 4 analyzes the computational complexity of the proposed algorithm. Furthermore, in order to reduce the complexity of the proposed algorithm, a simple but fast coding method is presented. Experimental results are presented in Section 5, followed by the conclusion in Section 6.

Section snippets

Background and problem statement

Since we want to jointly optimize the level and the map information of the residual coefficients in H.264/AVC, the coding process of the residual coefficients is introduced first.

Similar to the previous coding standard, H.264/AVC utilizes transform coding for the residual data, which are the errors between original pixels and predicted pixels. These residual data need to be transformed, quantized, and entropy coded in the encoder [11]. The transform method is the block based discrete cosine

Joint optimization of level and map information

Since we have known the fact that some of the nonzero coefficients in scanned arrays might not be valuable for encoding in RDO sense, it is necessary to design a coding scheme to jointly optimize the level and the map information of residual coefficients in H.264/AVC. In order to solve this problem, first, a linear hierarchical data group is defined in a scanned array for the purpose of algorithm design. Then we present a pruning algorithm to zero out the nonzero coefficients, which are not

Complexity analysis of proposed algorithm

The proposed algorithm contains two basic encoding modules: the module of pruning algorithm and the module of SQ selection for matching the pruning algorithm. The computational complexity of the two encoding modules will be introduced accordingly.

In the module of pruning algorithm, the competing framework for the selection of residual coefficients, which need to be pruned out, results in high complexity. The RD cost of each pruning case needs to be calculated in the competing framework; so the

Experimental results

Experiments have been performed on the JM12.4 reference implementation of H.264/AVC [16] and the proposed schemes. The codes are compiled using Visual C++ 6.0 on Windows XP (SP2) platform. The experimental results are achieved using computer with 2.59 GHz Pentium Dual-Core CPU and 0.98 GB memory. Ten video sequences shown in Table 1 have been tested in our experiments and the experimental results of the ten sequences have been given as follows.

Conclusion

In this paper, a joint optimization algorithm for the level and the map information of residual coefficients is presented in the H.264/AVC video coding. The proposed algorithm is composed of two encoding modules. One is the residual coefficients pruning module and the other is the module of scalar quantization step selection. To some extent, the proposed algorithm is similar to the space-frequency coding scheme. They all optimize the level and the map information by pruning out some

Acknowledgment

This work was supported by 2007 National Natural Science Foundation of China (Project no. 60602024) and 2009 Natural Science Foundation of Xi'an Jiaotong University (Project no. xjj2009042).

References (16)

  • T. Wiegand, G.J. Sullivan, A. Luthra, Draft ITU-T Rec. H.264/ISO/IEC 14496-10 AVC, Presented at the JVT ISO/IEC MPEG,...
  • Z. Xiong et al.

    Space-frequency quantization for wavelet image coding

    IEEE Trans. Image Process.

    (1997)
  • Z. Xiong et al.

    Wavelet packet image coding using space-frequency quantization

    IEEE Trans. Image Process.

    (1998)
  • D. Gleich et al.

    Progressive space frequency quantization for sar data compression

    IEEE Trans. Geosci. Remote Sensing

    (2002)
  • V. Velisavljevic et al.

    Space-frequency quantization for image compression with directionlets

    IEEE Trans. Image Process.

    (2007)
  • A. Shoa et al.

    Optimized atom position and coefficient coding for matching pursuit-based image compression

    IEEE Trans. Image Process.

    (2009)
  • T. Wiegand et al.

    Rate-constrained coder control and comparison of video coding standards

    IEEE Trans. Circuits Syst. Video Technol.

    (2003)
  • E.-H. Yang et al.

    Rate distortion optimization for H.264 inter-frame coding: a general framework and algorithms

    IEEE Trans. Image Process.

    (2007)
There are more references available in the full text version of this article.

Cited by (0)

View full text