Research On Screen Content Coding Method

With the rapid growth of cloud computing and the widespread promotion of the application such as virtual desktop,desktop cloud,and etc.,the screen content has become an integral part of the new generation of cloud-mobile computing model.Under this circumstance,it is much important to investigate screen content coding methods with high compression efficiency,good real-time performance,and moderate computational complexity,which are eagerly needed in practical applications as well.Nevertheless,since the statistics of screen content image/video is different from that of the natural image/video,it is often difficult to obtain satisfying compression efficiency using conventional coding methods.First,on introducing the statistics of screen content image,this study discusses typical spatial-domain coding methods of discontinuous image.State-of-the-art methods are classified into six classes,namely palette-index map based methods,template matching based methods,block matching based methods,dictionary-based methods,shape representation based methods,as well as chroma component coding methods.The hybrid framework of screen content coding is further summarized.This study also compares,analyzes,and discusses the advantages vs disadvantages of various methods.Second,this study discusses the screen content coding methods based on palette-index map,and introduces the index map coding method based on multi-stage prediction.Aiming at the low efficiency of index map prediction,this study analyzes the local directional correlation of screen contents through experiments.Both a template with 2?3 pixels for initial prediction and a template with 3?4 pixels for secondary prediction are then designed.Subsequently,this study puts forward a fast index map prediction algorithm based on the local directional correlation.Third,this study examines the influence of index assignment optimization on the index map prediction efficiency.Aiming at the index assignment and prediction of the pixel values within overlapped quantization zones of two base colors,this study addresses a joint fast index assignment and index map directional prediction algorithm.It is able to simultaneously implement the index assignment and index map prediction within one scan.We carry out massive experiments on 19 standard test video sequences and 3 test images.Experimental results show that the prediction accuracy of our two algorithms achieves 95.43% and 96.29% respectively which increases by 2.48% and 3.16% on average than the typical multi-stage prediction algorithm,especially suitable for those videos containing text characters and complex scenes with a large number of geometric elements.Moreover,the computational complexity is significantly lower than that of MSP.these algorithms thus can satisfy well the requirement of screen image coding.