Multiclass Classification for Effective Mode Decision in High Efficiency Video Coding and Beyond

Project: Research

View graph of relations



High Efficiency Video Coding (HEVC) has attracted much attention due to theincreasing need of compressing large-scale video data for storage and streaming overnetworks. It doubles the compression efficiency at the same video quality compared to itspredecessor H.264/AVC standard, and demonstrates a great potential for High Definition(HD) video markets, such as IMAX movies, immersive video conferences, surveillance,ultra HDTV and 3D-TV.To achieve higher compression efficiency, HEVC divides video to a sequence of framesand subsequently partitions it into a number of smaller blocks size from 64x64 to 4x4for better compression efficiency. This kind of partitioning can generally refer to modedecision problems in HEVC.The mode decision problems aim to minimize the bit rate as well to maximize the videoquality based on a user designed rate distortion metric in HEVC, which are extremelycomplex and with high computational cost as they need to exhaust all the possiblechoices in order to come up with the optimal solution. Hand-crafted features andstatistical-based methods are commonly used but they can hardly adapt to rapid changeof the demands of today’s video especially with the dramatic increase of mode choices inHEVC and beyond.In this research, we propose a new unified framework based on learning which aims toresolve the mode decision problems in HEVC. Firstly, the mode decision processes inHEVC will be formulated as multiclass classification and recursive decision problems,and a machine learning based encoder framework and solutions will be developed, whichtend to address future extensions of the coding standards. Secondly, representativefeatures will be explored for the classification problems; feature selection and sampleselection will be investigated to reduce data redundancy. Learning schemes will bedeveloped to improve classification performance and adaptability in HEVC.This research will benefit large-scale video processing and boost the applications ofHEVC for video transcoding, codec chip design, internet video, video communication andbroadcasting, etc.


Project number9042322
Grant typeGRF
Effective start/end date1/01/1726/08/20

    Research areas

  • Image and Video Coding , High Efficiency Video Coding , , ,