Mode dependent loop filter for intra prediction coding in H.264/AVC

https://doi.org/10.1016/j.jvcir.2013.06.016Get rights and content

Highlights

  • We analyze the limitations of the deblocking loop filter (DLF) in H.264/AVC.

  • We investigate the intra modes combinations characteristics in H.264/AVC.

  • A novel adaptive loop filter was proposed by combining a bank of wiener filters.

  • The filter parameters are trained for different intra modes combinations.

Abstract

In this paper, a high performance and low complexity loop filter is proposed for intra prediction coding. Although the deblocking loop filter (DLF) has achieved outstanding performance on suppressing quantization noise, it also induces details information loss because of the smoothing operation. To achieve better restoration performance, we propose a filter set named mode dependent loop filter (MDLF) which adaptively select the filter coefficients according to various local characteristics. In the homogeneous areas, the task of the filter emphasizes on smoothing the noise. In the heterogeneous areas, the proposed filter concentrates on preserving the details. Based on the spatial correlation assumption and statistical analysis, the intra mode combination is used to classify the training samples with different local characteristics. Then the classical least mean square error framework is employed to solve the coefficients for the proposed filter set. In this way, a more efficient adaptive loop filter scheme can be achieved for specific intra mode combination. Experiment results show that the proposed loop filter achieves superior coding gains compared to the H.264/AVC High Profile. Furthermore, relative to QALF+DLF, a comparable performance also can be achieved by the proposed MDLF with far less complexity increase.

Introduction

During recent decades, the development of digital multimedia technology has greatly changed our lives. In this process, the researches in image/video understanding [1], [2] and coding [3], [4], [5], [6] provide us efficient tools to process the video data for kinds of applications. For video compression applications, most hybrid video coding strategies employ block-based transforms among their compression tools. Even in the latest standardization work of HEVC [7], this framework is also inherited. Coarse quantization for the transform coefficients inevitably introduces visual artifact of disturbing discontinuities at the block boundaries [8]. To remove the blocking artifacts, the deblocking loop filter (DLF) [8], [9] is adopted in H.264/AVC, and still remains in HEVC. In the DLF, a bank of fixed low-pass filter are adaptively selected to smooth the block boundaries along horizontal and vertical directions. Although the DLF has achieved outstanding performance both in subjective and objective quality, two constraints still restrict its capability in restoring the degraded image. On the one hand, the achievement of DLF depends on the smoothing operation, which induces detail information loss with respect to original image. On the other hand, fixed filter coefficient is insufficient to adapt to different local features in a frame.

A number of recent researches have provide us helpful ideas in improving the existing loop filter. Some works focus on analyzing the perceptual character of an image [10], [11], while the others concentrate on improving the performance of the conventional DLF scheme. As a well-known optimal linear filter to cope with the image degraded by Gaussian noise, Wiener filter has been widely applied in various image/video denoising applications. To restore the degraded image by compression, the adaptive loop filter (ALF) is designed by using Wiener filtering in [12], [13]. Although Wiener filter can provide global optimum for the training samples, its performance may degrade in some local regions. To prevent the local deterioration effects, many methods have been proposed, which can be divided into two categories. The first class aims to classify the training samples, and the second one focus on switching on or off the Wiener filter in local regions. In the first category, a unified loop filter (ULF) was proposed in [14] which divided the pixels into enhancement group and restoration group to design corresponding filters. The classification criteria is mainly based on boundary-strength (BS) and gradient analysis in DLF [8]. Unlike the sorting principle in ULF, Chen et al. proposed a classified quadtree-based adaptive loop filter (CQALF) in [15]. The pixels are classified on the basis of whether modified by the DLF or not. Then, the new training samples are generated from these two categories of pixels via a weight average scheme. To minimize the mean squared reconstruction error, a system composed of three adaptive filters is proposed in [16], and two classes of new training samples are introduced to accomplish a jointly coefficients estimation.

The second category methods focus on avoiding over-filtering in some local regions. In the block-based adaptive filter control (BAFC) method [17], variable block size scheme is adopted to control the bit overhead of BAFC flags. Furthermore, in [13], the quadtree-based adaptive loop filter (QALF) is proposed to efficiently signal the block partition and filtering decision information, where a bottom-up method is used for the rate-distortion optimization (RDO) [18] of quadtree data structure. During recent video coding standardization work, the joint filtering (QALF+DLF) scheme is firstly developed in key technical area (KTA) software [19], and has been employed in the latest HEVC software [20]. Even though an outperforming performance has been achieved by QALF, the increased computational complexity brings great challenge in the real-time applications. In all of the methods mentioned above, most works focused on improving the performance of QALF. It is still a challenge in designing a efficient scheme to trade off the performance and complexity of the adaptive loop filter.

Unlike previous methods which produced the coefficients with the same scheme on the whole of a frame, we design a filter set named mode dependent loop filter (MDLF) for different intra mode combinations. Since each intra mode combination is related to specific local characteristics, our proposed scheme can restore these local regions efficiently. By utilizing the high correlation of the local features among consecutive frames, we can transfer the filter coefficients of a frame to the subsequent frames, which significantly reduces the complexity of the proposed algorithm.

The remainder of this paper is organized as follows. We firstly analyze the degradation model under hybrid coding framework in Section 2. Then, the detail description of our proposed MDLF is presented in Section 3. Section 4 shows the experimental results. Finally, we draw the conclusion in Section 5.

Section snippets

Analysis of degradation model under hybrid coding framework

In most hybrid coding schemes, the DLF plays an important role in improving the reconstructed image quality. Although the quantization noise could be efficiently suppressed in this process, additional degradation effect is also induced through the smooth operation. To investigate this degradation process, we review the work of DLF. In addition, since Wiener filter is an well-known optimal linear restore filter to tackle the images degraded by compression, the principle of Wiener filter is also

Filter design

In this section, we start with several intuitionistic observations which provide helpful guidelines for introducing our proposed algorithm. Then a detailed description of the intra mode dependent classification criteria is followed. At last, we present the filter design and optimization of our proposed MDLF scheme in details.

Simulation results

To verify the performance of our proposed loop filter, the MDLF scheme is implemented on the VCEG KTA2.4r1 reference software [19] under common test conditions for the intra only simulation [28]. The quantization parameter setting is QP={22,27,32,37}. A wide range of resolutions are covered in the test sequences, including CIF, 720p and 1080p. For each sequence, the first 100 frames are encoded in our experiment. In view of the effect from the quantization step, another experiment is

Conclusion

In this paper, a novel loop filter is proposed to restore the image degraded by compression. To adapt to various local characteristics, a samples classification criteria is established based on intra mode combination. The optimal filter set is derived by employing LMSE framework for each intra mode combination. The simulation results demonstrate that the proposed scheme provides superior coding gains relative to H.264/AVC High Profile. Furthermore, our proposed MDLF achieves comparable

Acknowledgement

This work was partially supported by National Natural Science Foundation of China (No. 61179060), National High Technology Research and Development Program of China (863 Program, No. 2012AA011503) and the Fundamental Research Funds for the Central Universities (ZYGX2012YB007 and ZYGX2012J019).

References (29)

  • H. Li et al.

    Saliency model based face segmentation in head-and-shoulder video sequences

    J. Visual Commun. Image Representation

    (2008)
  • Z. Wei et al.

    An efficient intra mode selection algorithm for h.264 based on edge classification and rate-distortion estimation

    Signal Process.: Image Commun.

    (2008)
  • H. Li et al.

    Adaptive scene-detection algorithm for vbr video stream

    IEEE Transactions on Multimedia

    (2004)
  • J. You et al.

    A multiple visual models based perceptive framework for multilevel video summarization

    IEEE Transactions on Circuits and Systems for Video Technology

    (2007)
  • ITU-T and ISO/IEC JTC 1, Generic coding of moving pictures and associated audio information—Part2: Video, ITU-T...
  • ISO/IEC 14496–2 (MPEG-4 Visual Version 1), Coding of audio-visual objects—Part 2: Visual (Apr....
  • ITU-T, Video coding for low bit rate communication, ITU-T Recommendation H.263 version 1...
  • ITU-T Recommendation H.264 and ISO/IEC 14496–10 (MPEG-4) AVC, Advanced Video Coding for Generic Audiovisual Services...
  • Joint Collaborative Team on Video Coding, High efficiency video coding (Hevc) Text specification draft 6, JCTVC-H1003,...
  • P. List et al.

    Adaptive deblocking filter

    IEEE Trans. Circuits Syst. Video Technol.

    (2003)
  • T. Wiegand et al.

    Overview of the h.264/avc video coding standard

    IEEE Trans. Circuits Syst. Video Technol.

    (2003)
  • H. Li et al.

    A co-saliency model of image pairs

    IEEE Trans. Image Process.

    (2011)
  • T. Chujoh, A. Tanizawa, T. Yamakage, Adaptive loop filter for improving coding efficiency, ITU-T SG16 Contribution,...
  • T. Chujoh, N. Wada, G. Yasuda, Quadtree-based adaptive loop filter, ITU-T SG16 Contribution, C181, Geneva, Switzerland,...
  • Cited by (1)

    View full text