| With the improvement of living standards,people have higher and higher requirements for visual experience,especially in the field of visual entertainment such as movies,games and television,people have higher requirements for the details of pictures,the brightness of colors,etc.And all of these require higher dynamic range to support.Therefore,the research and application of high dynamic range(HDR)video reconstruction has become a research hotspot in the field of visual art.However,the reconstruction of HDR video is a very complex problem.The existing methods either do not consider the constrained solution space,or simply reverse the camera imaging pipeline,without directly constructing a unified HDR imaging process.At the same time,the current HDR video reconstruction work does not consider a practical problem,that is,a large number of standard dynamic range(SDR)videos are compressed to meet the needs of saving storage space and so on.In addition,due to the existing research schemes focus on the reconstruction of SDR to HDR video,and a problem of practical significance,the quality enhancement of compressed domain HDR to HDR video has not been given much attention.And these are all very important problems in the process of HDR video reconstruction.Based on the above problems in HDR video reconstruction,this paper focuses on the camera imaging pipeline to study the problems in HDR video reconstruction from multiple perspectives and those ignored by researchers.This paper first investigates the related schemes of HDR video reconstruction at home and abroad.Then,in view of the shortcomings of existing schemes and the lack of current research,three related research work is carried out,which is summarized as follows:1.This paper analyzes the mathematical formulas of the camera imaging pipeline and derives the formulas of HDR image imaging.Then,this paper uses convolutional neural networks to simulate the formulas to reconstruct HDR images/videos.At the same time,the skip connection structure in the UNet network is corrected to adapt to the HDR problem.Compared with the previous methods,this method can reduce the solution space of HDR image/video reconstruction problems,and generate more realistic HDR images/videos.2.This paper presents and solves a new problem,namely compressed domain HDR to HDR video reconstruction.By analyzing the SDR video compression scheme,this paper integrates the prior knowledge of the compressed video into the mathematical formulas of the HDR image reconstruction,so as to improve the quality of the reconstructed HDR video.3.This paper proposes a new research direction of practical importance,the quality enhancement of compressed domain HDR video.In order to promote the exploration of this problem by researchers,this paper proposes the first compressed domain HDR video quality enhancement dataset,and builds the corresponding deep learning model based on this dataset.This model can improve the quality of compressed domain HDR videos,so as to save the bit rate in the transmission of HDR videos. |
