Novel distribution model of transformed coefficients in video coding using quad-tree structured block partitioning

• Published in: Multidimensional systems and signal processing Vol. 28; no. 4; pp. 1589 - 1609
• Main Author: Kang, Je-Won
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.10.2017; Springer Nature B.V
• Subjects: Artificial Intelligence; Circuits and Systems; Coding; Coding standards; Coefficients; Density; Electrical Engineering; Engineering; Mathematical models; Parameter estimation; Partitioning; Signal,Image and Speech Processing; Video compression; Parameter estimation; HEVC; Quad-tree structure; R-D model; DCT coefficient distribution; Rate control
• ISSN: 0923-6082; 1573-0824
• DOI: 10.1007/s11045-016-0438-8

Today’s video coding standard such as high efficiency video coding uses a full quad-tree structured block partitioning, so the underlying statistics of transformed coefficients becomes more complicated to estimate than the previous standards due to the coding structure. However, a statistical distribution of transformed residue is important for a design of a smart encoder. Thus, in this paper, we present a theoretic analysis of a distribution of transformed coefficients produced from an encoder using different transform sizes, and derive a probability density function (pdf) for the estimation. The proposed density model provides a more accurate distribution model than the conventional pdfs. Parameters are theoretically estimated, and rate-distortion model is established from the proposed pdf. We also apply the proposed method to a rate control problem to show the efficiency of the proposed density model. Our experimental results show that the proposed method is better capable of modeling the mixed sources of multiple-type transform coefficients occurred from the quad-tree coding structure of transform and provides an accurate estimate in rate control.