A fast and HEVC-compatible perceptual video coding scheme using a transform-domain Multi-Channel JND model

Gang WANG, Yongfei ZHANG*, Bo LI, Rui FAN, Mingliang ZHOU

*Corresponding author for this work

Research output: Journal PublicationsJournal Article (refereed)peer-review

4 Citations (Scopus)

Abstract

Coding optimization methods incorporating the just noticeable distortion (JND) model, called perceptual video coding (PVC), have drawn much attention in recent years for better video coding performance. To further remove perceptual redundancy in every channel and improve the coding performance, this paper proposes a fast PVC scheme in the latest High Efficiency Video Coding (HEVC) framework based on our proposed variable block-size transform-domain multi-channel JND model. Firstly, through extensive experiments, we find out for the first time that the contrast masking (CM) effects for chroma channels show a lowpass property in frequency, which differs from the luma channel that has a bypass property. Based on this observation, CM effects in chroma blue (Cb) and chroma red (Cr) channels are modeled as a continuous function for variable-sized blocks, respectively. Secondly, since different characteristics of the human visual system (HVS) exhibit quite distinct effects in luma and chroma channels and effects in chroma channels were not well explored, we develop a new JND model through comprehensive consideration for both luma and chroma channels of five typical HVS effects, with especial focus on parameterized modeling of chroma channels in each effect. Finally, to incorporate the proposed JND model into the latest HEVC coding framework, a multi-channel coefficients suppression method based on JND thresholds and quantization parameter (QP) ranges is proposed in the transform and quantization process, which can decrease the computational complexity. Extensive experimental results show that the proposed PVC scheme implemented in HEVC reference software (HM15.0) can yields significant bit saving of up to 25.91% and on average 9.42% with similar subjective quality, compared to HM15.0, and consistently outperforms two PVC schemes with much reduced bitrate and complexity overhead.

Original languageEnglish
Pages (from-to)12777-12803
Number of pages27
JournalMultimedia Tools and Applications
Volume77
Issue number10
Early online date4 Jul 2017
DOIs
Publication statusPublished - May 2018
Externally publishedYes

Bibliographical note

Publisher Copyright:
© 2017, Springer Science+Business Media, LLC.

Keywords

  • High efficiency video coding (HEVC)
  • Just noticeable distortion (JND)
  • Perceptual video coding (PVC)

Fingerprint

Dive into the research topics of 'A fast and HEVC-compatible perceptual video coding scheme using a transform-domain Multi-Channel JND model'. Together they form a unique fingerprint.

Cite this