Research On Multi-Laver-Based Video Coding

Multi-layer coding structure is the structure which encodes video into separable layers and employs inter-layer prediction to exploit their correlation.The traditional multi-layer coding structure is applied in spatial scalability and quality scalability of scalable video coding,in which each frame in enhancement layer has a corresponding frame with lower resolution or quality in base layer.In this paper,a new multi-layer coding structure is proposed,the enhancement layer is the encoding video while the base layer is just composed of several images with common information.With the new multi-layer coding structure,the overall coding efficiency is higher than single layer video coding.In other words,the new multi-layer coding structure can be used for higher efficiency video coding.In this paper,we research multi-layer-based video coding and make the following contributions.1.Propose a new multi-layer coding structure based on knowledge libraryIn traditional multi-layer coding structure,there is a consistent one-to-one match between frames in enhancement layer and base layer.In contrast,the new multi-layer coding structure is based on video analysis.The base layer is generated by extracting several common images in encoding video.Then inter-layer prediction is employed to exploit the inter-layer correlation so that the overall coding efficiency is higher than single layer.The frames in base layer need to be stored for long to provide reference for enhancement layer.So in this paper,a knowledge-library-based video coding framework is proposed.We solve two questions under the proposed framework.First,we research the way to build the knowledge library base layer and propose two methods.Method 1:Extract the scene change frame and random access frames of each scene as key frames.Classify the key frames according to their similarity and eliminated frames belonging to repeated scenes.Finally a representative frame is chosen from each scene as library frame.This method can assure the high correlation between base layer and enhancement layer and improve the coding efficiency.Method 2:Extract the scene change frames of each scene and use SIFT-based repeated scene detection to eliminate frames from repeated scenes.The initial library frame candidates are derived.Then extract more additional candidates from each scene according to accumulated scene content change.This method can work in synchronization with video coding and is suitable for real time application.Second,we propose the coding method under the new multi-layer coding structure.To improve video coding efficiency and maintain the random access functionality,the library frames are intra coded and the library frames is cleared by random access point.In enhancement layer,each random access segment is coded by referencing the most similar library frame.The most similar library frame is searched using color histogram difference as similarity measurement,which is simple and can ensure the prediction efficiency of the library frame.2.Research on coding methods under traditional multi-layer coding structureWe investigate the inter-layer prediction techniques in traditional multi-layer coding structure,including inter-layer intra mode prediction and inter-layer motion prediction.With these two techniques,the inter-layer correlation of intra mode and motion information can be well exploited.The coding efficiency of enhancement layer can be improved while the complexity is reduced.