Optimization For Cross Random Access Point Reference In Video Codec

In applications that support random access,one video is divided into multiple independently decoded random access segments by multiple random access points.Therefore,the video frames in a random access segment cannot reference to the video frames in other random access segments,which means that the temporal correlation among similar random access segments cannot be effectively used.Some researchers propose a video coding technology called cross random access point reference,which can greatly improve video coding performance by using the correlation among similar random access segments.The main goal of previous work is to explore the theory of this technology.In order to promote the practical application of this technology,in this paper,we research and optimize the specific implementation details of the cross random access point reference involving the following aspects: the problems in the inter prediction when the original cross random access point reference is put into practical application,as well as the problem of how to correctly decode the video bitsream that uses the cross random access point reference and how to continuously display the reconstructed video frame at a decoding end with limited resources.Firstly,for the problems in the inter prediction when the original cross random access point reference is put into practical application,we propose a syntax elements based library pictures indexing scheme,by using which the information needed for the index of library pictures can be obtained from the bitstream and the correct referenced library picture can be got correspondingly when random access.And we propose a motion vector scaling scheme based on the content characteristics of library pictures.By reasonably estimating the time-domain distance of the motion vector pointing to a library picture according to the characteristics of the content of library pictures,the correlation between the motion vector pointing to a library picture and the motion vector pointing to a video frame can be effectively utilized through time-domain scaling,which achieves a coding gain of 0.26% on the AVS3 test sequence.Secondly,for decoding end with limited resources existing in the actual application scenario,we propose a bitstream organization and decoding scheme to correctly decode the video bitstream that uses the cross random access point reference.By inserting the library bitstream into the video bitstream to obtain a merged bitstream,as well as designing the corresponding bitstream organization method and the decoding scheme for the merged bitstream,the decoding end that has only one decoder can also correctly decode the library picture bitstream and video picture bistream in the merged bitstream to get the correct reconstructed video.This enhances the robustness and applicability of the cross random access point reference video decoding.Finally,in order to ensure that a decoder that complies with video coding standards can decode the merged bitstream and display the reconstructed video in a timely manner,we propose a non-freeze timing model for decoding and displaying the merged bitstream.We improve the original timing model for decoding and displaying limitedly improving the decoding frame rate based on a library bitstream limitedly insertion frequency constraint and delaying the initial display time of the reconstructed video,which makes that the decoding finish time of each video frame in the merged bitstream can be guaranteed to be earlier than its display start time.And by correctly simulating the input and output process of the bitstream buffer of the decoder with the improved timing model for decoding and displaying,we propose a method for constraining the bitrate variations of the merged bitstream.Through the research and optimization of the above three aspects,we complete the implementation scheme of the decoding end to support the cross random access point reference with sufficient and insufficient resources.After adding the above improvements,the cross random access point reference video codec technology has been adopted by the domestic new-generation video coding standard AVS3,which promotes the cross random access point reference video codec technology to be put into practical industrial applications.