High-performance Video Codecs Based On Multi-core DSP Processors Technical Research

Video encoding,also called video compression,is an important research direction in the field of image processing in the information age.Video coding means using the correlation of video data to remove the redundancy in the image information.The redundancy mainly includes the information redundancy of the image spatial correlation,the time correlation and the frequency domain correlation,and the statistical redundancy of the coded symbols.According to whether the compression algorithm produces distortion,video compression is classified into lossy compression and lossless compression.With the continuous improvement of video resolution,frame rate,and image depth,traditional coding standards have been unable to meet the demand for high compression rates.As a new coding standard,H.265/HEVC greatly improves the video coding efficiency,but at the same time it also brings an increase of tens of times the computational complexity,which results in the coding speed being too slow.There are two main directions for video coding research.One is to improve image coding performance by fully exploiting correlation and remove information redundancy,and the other is to increase coding speed meanwhile ensuring coding quality.The main purpose of this paper is to improve the encoding speed by optimizing the encoding algorithm and using the advantages of the multi-core DSP processor,and to design and implement a video encoding and decoding system that can be practically applied.Intra prediction encoding is mainly to remove redundant information in space.By traversing 35 prediction modes and calculating the Lagrangian rate distortion cost,the optimal prediction mode is selected.The selection of the optimal prediction mode is divided into two processes: rough selection and fine selection.Although the Hadamard transform replaces the discrete cosine transform and the discrete sine transform to improve the calculation speed in the rough selection process,its essence is still an exhaustive algorithm.This article uses the feature of most images or most of the area of an image is flat to optimize the rough selection process by counting the number of times and probabilities that 35 prediction modes were finally selected as the optimal prediction mode,then grouping the prediction mode,classifying within the group,and allocating different computing resources to different prediction modes with different probabilities.The results show that under the premise of ensuring the image quality,the encoding speed of I picture can be effectively improved.Compared to x265,the speed increases by about 10% under the same conditions.Multi-core DSP processors have great advantages in video coding.This paper selects the C6678 multi-core DSP as the platform to implement a single-core video encoder that conforms to the H.265/HEVC standard.In order to improve the encoding speed and take advantage of multi-core processors,a multi-core parallel encoder that requests and distributes in GOP is designed and implemented.Due to the mutual independence between closed GOP,parallelism based on GOP can reduce synchronization and coordination among multiple cores.By storing the video data in the multi-core shared DDR to avoid transferring large amounts of data between cores,and simultaneously storing the data in banks,the waste of CPU cycles caused by multi-core competing shared memory is reduced.In the encoding and decoding system implementation,the image buffering mechanism of double-send and double-receive buffers is designed to realize the asynchronous operation of transmission and encoding,and the flow control is performed through the request response mechanism and data package encapsulation.The test results show that multi-core DSPs have superior performance in video coding.For SD video,the encoding speed can reach 25 fps or more and can be applied to actual codec scenarios.