High Image Compression Of RMB Image Based On HEVC

RMB is the legal tender of our country.With the development of social economy,the demand for RMB is increasing.At the same time,the whole society demands more and more quality of RMB.According to the technological requirements of China Banknote Printing and Minting Corporation,it is necessary to store the printed RMB images.However,due to the large amount of data,the production cost of enterprises is directly increased.The goal of this project is to reduce the storage space of RMB image by HEVC(High Efficiency Video Coding).In the case of guaranteeing the quality of pictures,the operation time is greatly reduced.For enterprises to reduce production costs and labor costs.Although the HEVC official reference software HM provides standard encoding,basic configuration parameters can be modified.However,if you want to implement a more optimized algorithm based on HEVC,you need to modify the HM source code to replace or supplement one of the steps.At the same time,according to the characteristics of RMB image and its practical application in production,the optimization should be targeted.In this paper,we propose two optimization schemes,namely,TZ(test tone)optimization algorithm and CU(Coding Unit)block selection based on lifting tree.Optimize TZ algorithm: it is optimized from traditional TZ algorithm.The optimized TZ algorithm first uses the MV(Motion Vector)prediction method to find the initial search point for each PU(Prediction Unit).Then perform circular search of traditional TZ in parallel for each PU.Change the termination condition to the traditional TZ termination condition for all PUs at the same time,that is,for all PUs,the best advantage of the two adjacent steps is the same.Optimization of TZ successfully solves the problem of not searching across regions and escaping from local extremes.Experiments show that the optimized TZ algorithm reduces the average time by 24%,improves the compression rate by 2.4% and reduces the distortion rate by 0.02% compared with the traditional TZ algorithm.CU block selection based on lifting tree: it is a specific algorithm for video sequences in RMB.Because of the training set,the prediction set is RMB video sequence,and the ensemble learning methods such as lifting tree are better for the problem of feature duplication.From more than 20 parameters to be selected,the lifting tree algorithm is implemented to extract the four features of CU block partition,merge residue of the current block,which is close to the left,upper left and upper right.After extracting features,the focus is to decide whether to perform depth + 1 operation in partitioning CU blocks.Compared with HM,it requires top-down iterations to partition the CU blocks,lifting the tree faster and ultimately reducing the compression time by 35%.At the same time,the loss of distortion rate and compression rate is very small.It only reduced the compression rate of 0.05% and increased the distortion rate of 0.09%.Finally,the optimized TZ algorithm and the lifting tree algorithm are applied to the reference software HM,and the video sequences generated from 10 groups of 1000 RMB images are tested.The experimental results show that the average compression time is reduced by 51%,the compression rate is increased by 0.05%,and the distortion rate is only increased by 1%.