Research On Fast Intra Coding Algorithm In Versatile Video Coding Standard

In recent years,with the development of the market demand for high-definition videos,the compression requirement for high-definition videos is also getting stricter and stricter.The latest video coding standard Versatile Video Coding(VVC)is proposed under this background.VVC still follows the traditional hybrid video coding framework.On the basic of the video coding standard High Efficiency Video Coding(HEVC),many new technologies have been added into the key modules of VVC to improve compression performance.But the coding complexity of VVC has increased dramatically,which resulting in too long encoding time and making its practical application more difficult.Therefore,reducing the coding complexity while maintaining the coding performance of VVC is a current research hotspot in the field of video coding.This paper mainly studies the fast algorithms of intra prediction in VVC,focusing on the research on intra mode selection and the secondary transform selection.Compared with HEVC's 35 intra prediction modes,VVC increases the number of intra prediction modes to 67 for higher coding performance.In the transform module,VVC adopts the low-frequency nonseparable secondary transform(LFNST)technology,which adds two different kinds of transform kernels to perform the secondary transform on the basic of the primary transform results.Therefore,the intra prediction process of VVC is more complicated.Aiming at the high coding complexity of intra mode selection and the secondary transform kernel selection,this paper proposes a fast VVC intra prediction algorithm based on the existing encoded and correlation information during the intra prediction process to achieve fast intra mode selection and fast LFNST kernel selection.Firstly,in order to reduce the complexity of intra prediction mode selection process,the existing encoded information for statistical analysis is utilized to implement a fast algorithm for this process.On the one hand,by analyzing the relationship between the modes after rough mode decision(RMD)process and the most probable modes(MPMs),the number of modes in the candidate list is reduced to reduce the complexity of it.On the other hand,the similarity between the RMD cost and the rate distortion optimization(RDO)cost is utilized to implement the early termination of the RDO process and further optimize the intra prediction process.Experimental results show that the proposed algorithms can reduce the encoding time by 30.61% on average while ensuring the same encoding quality comparing to VVC reference software VTM2.0,which outperforms the state-of-the-art algorithm.Secondly,in order to reduce the complexity of intra prediction secondary transform selection,the existing encoded and correlation information are used to implement a fast secondary transform selection algorithm.On the one hand,this paper proposes to utilize the correlation between the current coding unit(CU)and its related CUs' best transform selection to determine the best transform selection of the current CU in advance,and terminate the secondary transform selection process early.On the other hand,by analyzing the influence of the number of non-zero primary transform coefficients on the best transform selection,a threshold-based termination algorithm for LFNST is proposed.Experimental results show that the proposed algorithms can reduce the encoding time by 18.10% averagely while ensuring the similar encoding quality comparing to VTM6.2.Finally,a joint simulation is performed on the intra prediction fast mode selection algorithm and the fast secondary transform selection algorithm,which demonstrates that the two key processes of intra prediction---the intra mode selection and the secondary transform selection have a great influence on encoding time.Experimental results show that the coding time is reduced by an average of 34.38% when the bit rate is not significantly increased and the peak signal to noise ratio(PSNR)is basically unchanged comparing to VTM6.2.