Research On High Dynamic Range Tone Mapping Algorithm Based On Perceptual Optimization

With the increasing development of multimedia content on the Internet,the content is becoming more and more varied and of higher quality.Electronic dis-play devices play an indispensable role in human life.As receivers of digital infor-mation,humans are constantly acquiring content of interest from the vast amount of digital information available to them.In recent years,with the advancement of various capture devices such as video and photography,people have become in-creasingly interested in the pursuit of a high quality viewing experience.While the usual capture devices are already capable of capturing high dynamic images of the real world,high quality display devices are not yet widely available due to technical problems,and common personal display devices are still ordinary monitors,which can only display a limited dynamic range.How to display high quality and high dynamic range images or videos on a limited dynamic range display device,and retain richer details and contrast,as well as more realistic becomes an urgent problem to be solved,in academic circles,this operation is called tone mapping,now become a hot topic of academic research.A large number of excellent algorithms have been proposed around tone mapping in the last two decades.However,the existing algorithms still face many problems,such as extra noise,halos and loss of detail in the tone mapping results,and most of them lack more robustness for a wide range of high dynamic range content,and a large number of algorithms require high precision computation,which is extremely time consuming and computationally expensive.Therefore,this paper proposes an efficient tone mapping model to address these issues.The main research in this paper consists of the following two aspects.(1)A neural network model is proposed to address the problem that existing tone-mapping algorithms are extremely computationally expensive in terms of time and lack of robustness.Previous tone-mapping algorithms are usually conventional algorithms,divided into two categories: local algorithms and global algorithms.Global algorithms operate uniformly on the whole image and are therefore time-costly and less computationally expensive,but produce results that are often ac-companied by a lot of noise and lack generalisability for use in the real industry.Local algorithms usually divide the original high dynamic range image into a base layer and a detail layer,and process the two parts separately,thus producing better and more robust results than global algorithms,but usually taking longer.The net-work model designed in this paper takes into account the advantages of both global and local algorithms from the human visual perception system,and uses the most streamlined network model to collect a large and diverse set of high dynamic range content for training.(2)To address the lack of interpretability of existing algorithms and the lack of consideration of real-world luminance,this paper considers the absolute luminance range of the real world for the first time.Due to the complexity of real-world lu-minance scenarios,existing high dynamic range images inherently have a relative luminance range that lacks calibration.Existing tone mapping algorithms usually ignore this important information and process the uncalibrated high dynamic range directly,producing erroneous viewing results in some extreme scenarios due to the lack of calibration of the high dynamic range images.In this paper,a tonal map-ping algorithm based on an adversarial neural network is proposed to address these problems by designing additional luminance prediction models in the network de-sign and using a generative adversarial neural network structure to produce more natural results with better subjective perception.The effectiveness of the proposed luminance prediction model is demonstrated experimentally and has the additional extension of detail enhancement due to the physical characteristics of luminance in the real world.