Low Frequency Domain Aided Texture Synthesis for Intra Prediction

  • Authors:

  • Xiaowei Sun;Baocai Yin;Yunhui Shi

  • Affiliations:

  • Multimedia and Intelligent Software Technology Beijing Municipal Key Lab, College of Computer Science and Technology, Beijing University of Technology, Beijing, China 100124;Multimedia and Intelligent Software Technology Beijing Municipal Key Lab, College of Computer Science and Technology, Beijing University of Technology, Beijing, China 100124;Multimedia and Intelligent Software Technology Beijing Municipal Key Lab, College of Computer Science and Technology, Beijing University of Technology, Beijing, China 100124

  • Venue:
  • AMT '09 Proceedings of the 5th International Conference on Active Media Technology
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

To save bit-rate, texture synthesis can be employed in video coding to generate similar texture instead of encoding the texture region. Without mean squared error (MSE) as the distortion criterion, high quality but different texture can be synthesized with few bits. One key problem that embarrasses the application of texture synthesis in video coding is the annoying artifacts. In this paper, we describe a new synthetic region selection strategy which can simultaneously save bit-rate and avoid the annoying artifacts. Our method is based on low frequency consistent. That is, a low resolution version of video sequence is coded and sent to the decoder side, and the texture synthesizer should strictly accord with the decoded low resolution picture. In the high resolution layer, structure and sampler blocks are identified and encoded using MSE as the distortion criterion. We implement our scheme in JSVM and the bit-rate saving account to 18% compared with SVC with the same subjective quality.