Perception Oriented HDR Video Coding With HEVC Main 10 Profile

Conventional Standard Dynamic Range(SDR)videos have been widely used in many applications.Main reason of its widespread adoption is that the compact representation as a byte is convenient for storage and transmission.The SDR videos can only cover a limited dynamic range and color gamut,while luminance levels and color gamut in the real-world can vary a lot more than what is captured by SDR imaging.High dynamic range(HDR)videos overcome the shortcomings of SDR videos by more successfully depicting the real world.Different from the SDR video which is represent by integer values,the HDR video is represent by float values.Thus,storage and transmission of the HDR content need much larger memory and bandwidth.The limitation of network bandwidth hinders the widespread adoption of HDR videos.Hence,it is necessary to realize efficient HDR video coding to reduce the bandwidth requirement.Due to the ultimate receiver of video signal is human eyes,we consider removing the visual redundancies by using the characteristics of human visual system(HVS)to improve the coding efficiency for HDR videos.The main work of the thesis is in the following aspects:1.A backward compatible opto-electrical transfer function(OETF)is proposed.Hybrid Log-Gamma(HLG)OETF has backward compatibility,and the second part of HLG is logarithmic mapping function.Uniform rational mapping function(URMF)is a fast and simple quantization technique which approximates a logarithmic mapping function using rational quantization.The URMF is more efficient than the logarithmic mapping function and it successfully meets the HVS characteristics.We replace the logarithmic mapping function of HLG with URMF to generate a new OETF.The proposed OETF is efficient and it can improve reconstructed HDR videos quality.2.A HEVC main 10 encoder optimization algorithm is proposed.We proposed a 10 bit free-energy based just noticeable distortion(FEJND)model based on the HVS.Then the 10 bit FEJND model is used to obtain the perceptual Lagrange multiplier for each coding unit(CU),and then performed perceptual CU splitting and block merging by minimizing rate-distortion(RD)costs.The proposed optimization algorithm achieves a significant bit-rate reduction and successfully improves the perceptual quality of the reconstructed HDR videos.3.A further optimization algorithm for the HEVC main 10 encoder is proposed.The calculation of the FEJND is a very time-consuming process which increases the computer complexity of HEVC main 10 encoder.To overcome shortcomings of FEJND model,we designed a visual regularity based JND(VRJND)model.Due to the VRJND model has low computational complexity and it can represent HVS more accurate.We use VRJND model to guide the CU splitting and merging.The VRJND based optimization algorithm improves the coding efficiency and has low computational cost.4.A fast CU size decision algorithm is proposed.The visual regularity model can not only be used to describe the HVS,but also can be used to represent the structural information of the image.After analyzing the relationships between the optimal CU size and visual regularity model,we designed an early termination algorithm and prediction algorithm for fast CU size decision.This fast CU size decision algorithm is a general algorithm,which means it can be used in HEVC main profile and HEVC main 10 profile.