Research Of HEVC Optimization Based On Texture Features And Visual Attention

With the rapid advance of the Internet and the popularity of video applications,the video quality is desired higher and higher.High-quality videos require huge volumes of data to describe the details of the picture resulting in a surge in video data.High Efficiency Video Coding(HEVC)is the new generation coding standard for high-resolution videos,whose core goal is to double the video compression efficiency on the basis of H.264/AVC High Profile.However,the improvement of its compression efficiency also brings high computational complexities,which seriously affect the HEVC promotion and application.Textures of the objects in videos are expressed through local pixels’ arrangements and changes,which are usually in slow or cyclical change and have certain regularity.Coding Unit(CU)is the basis unit of the video coding in HEVC.The areas of simple texture are divided into large-size CU with low depth level,and the areas of complex texture are divided into small-size CU with high depth level.Besides,the divided depth of CU will be same if the area has similar texture.However,the algorithm of CU division has high computation and is one of the main factors of restricting the HEVC performance.Therefore,considering the texture feature of the video in HEVC,the depth of the CU can be predicted and the coding time can be reduced effectively.On the other hand,the eye is the ultimate receptor of the video signal,which means the video quality equals the subjective perceptions quality of our eyes.Our Human Visual System(HVS)does not pay equal attention to all the areas of the video.When reallocating the rate resource according to the visual attention,the visual redundant can be removed effectively,and compression performance can be improved highly.Therefore,researching on HEVC based on texture feature and visual attention has important theoretical significance and broad application value.Based on the study of CU division principle,a fast algorithm of CU division based on Canny operator was proposed,though which the CU was advanced into sub-division,coding complexity was reduced and the coding process was speeded up.Then,the visual attention model was established according to the characteristics of HVS.Based on the attention of the Largest Coding Unit(LCU),the adaptive coding compression algorithm was established to reallocate the rate resources of different attention areas,through which the video coding rate was reduced under the premise of high visual quality and the overall compression ratio was improved effectively.The main contents of this thesis include the following three aspects:(1)Research on depth prediction of CU to optimize the HEVC.Firstly,the correlation of the CU depth between its neighborhood and reference frame at the same position was researched.The mathematical relationship between coding depth and the distribution of the texture was deduced.Then,the Canny operator was introduced to separate the texture area of the key frame.Finally,the initial depth of CU was predicted according to the texture distribution.Results experienced on the HM16.6 suggest that the algorithm has high practical utility in digital animations.(2)Simulate the selective attention mechanism of HVS to establish the visual attention model.According to the visual selective attention,the motility factor,texture complexity factor,contrast factor and relative brightness factor were introduced to establish visual attention model.In order to keep the computational complexity at low level,the Gray Projection Method(GPM)was adopted to calculate motility factor,which has low complexity and strong robustness.Besides,the contrast factor is acquired based on Pixel Four Neighbors(PFN).The relative brightness factor is obtained according to the Coding Four Neighbors(CFN).(3)Adjust the resources of bit rate according to the different attention of CU to achieve the self-adaptive coding compression.According to the characters that our eyes are more concern about the structure distortion rather than pixel distortion,for the high attention LCUs,the Structural Similarity Index Measurement(SSIM)distortion optimum algorithm which is constructed in this thesis was applied instead of the traditional error square sum algorithm.In the low attention LCUs,the Lagrangian multiplier was adjusted based on visual attention to coarsely quantify these areas and achieve the effect of improving compression radio and reducing the bit rate.