Research On Tone Mapping Algorithm For High Dynamic Range Images

Although the concept of high dynamic range(HDR)is not well known to many people,HDR has appeared in most people's mobile phones,televisions,digital cameras,drones,cars and so on.The dynamic range is often expressed as the ratio of the maximum and minimum values of the brightness values of the scene or image.At present,image capturing devices and display devices commonly used are in the category of low dynamic range(LDR),but the dynamic range of nature and human eyes is far more than this.Underexposure or overexposure in the image tends to lose a lot of detail when the HDR scene is captured and displayed on the LDR device.How to compress the dynamic range of the HDR scene to accurately display it on the LDR device--tone mapping,is the main research direction of this paper.Research on HDR-related technologies is essential in the areas of transportation,security,entertainment and artificial intelligence.Based on my internship experience,we first put forward the research background and research purposes,and expound the significance of the research.Then the imaging theory is introduced from the aspects of human eye structure and camera structure.At the same time,through the research of industry product reports,product launches and All in Camera of Zhihu Column,we introduce two main ways to obtain HDR images at present,sensor HDR and post-processing HDR.Then we introduce the previous research on the theory of tone mapping of HDR images,analyze the previous global tone mapping method and local tone mapping method,and enumerate the classic tone mapping algorithms.Inspired by the naturalness of the image and the more sensitive characteristics of the human eye to local features,we propose a tone mapping algorithm based on normal distribution.By analyzing the shortcomings of Meylan's retina model algorithm,this paper proposes an improved algorithm based on mean-pooling and bilinear interpolation,and then proposes a new global tone mapping algorithm.This paper proposes a tone mapping method based on normal distribution,including global tone mapping algorithm and local tone mapping algorithm.Global tone mapping adjusts the overall brightness of the image by changing the image histogram distribution,making the human eye feel more natural.Local tone mapping applies a global tone mapping algorithm to the image local area.A rectangular area centered on one pixel is used as a sliding window.We change the histogram of the window by counting the mean and variance of the window pixels.The histogram in the window is matched to the shape of the normal distribution function with u as the mean and ? as the standard deviation.The algorithm traverses each pixel in the image,applies a method of global tone mapping based on normal distribution to the window,and saves the mapped value of the pixel.The local tone mapping method based on normal distribution can compress the global contrast well.The tone-mapped image has rich local detail and local contrast,which is consistent with the human eye's sensitivity to local features,while the image retains its natural appearance.In addition,after the acceleration of OpenCL,the algorithm operation time is shortened a lot.The improved retina model algorithm in this paper can achieve good contrast and solve the halo problem.The global tone mapping algorithm proposed on this basis can also obtain similar results,and its time complexity is O(1).The algorithm can be applied to compression of HDR images and enhancement of ordinary images.The algorithm can effectively improve the quality of images captured by cameras such as traffic and surveillance.The algorithm can be applied to scenes such as object recognition,face detection,and self-driving cars to improve their anti-illumination performance.