Research On CRF Estimation And HDR Imaging Method Based On Single Image

With the extensive application of digital image processing technology in modern information science,high dynamic range(HDR)imaging method has become one of the hotspots in image processing and computer graphics in recent years.Image is the main source of information for human visual systems.HDR imaging turns to break through the limitation of dynamic range captured by sensors and enhance the color resolution and distinguishable range of visible light,presenting more gentle tone variation and more realistic visual effects.The main contents of this thesis include camera response function(CRF)estimation based on single frame image and HDR image generation.Furthermore,mainstream methods of HDR imaging are described in detail and analyzed experimentally.The contrast and exposure of a image are greatly related to the properties of CRF.Taking full adavantage of CRF characteristics,the image contrast or exposure can be enhanced or adjusted,and thus significantly improve the visual effects of the image.Unlike the multi-exposure HDR imaging technique,we will obtain the inverse camera response function by the nonlinearity properties of edge pixel's color in single image,and correct the the original low dynamic range(LDR)image by the inverse CRF to generate the HDR image.Compared with the previous work,the algorithm is improved in the following aspects: Firstly,more stringent constraints are imposed in the process of extracting effective edge path information,which improves the prediction accuracy of inverse CRF.Secondly,using the nonparametric estimation method based on a window function to calculate the prior probability value of the inverse CRF.Thus,the hypothesis of the distribution of the prior probability model is weakened.On the basis of above work,this thesis presents a HDR imaging algorithm based on CRF estimation using single LDR image.The algorithm uses local weighted linear regression to calculate the original scene radiance for each pixel in the LDR image.Thus,the sample distribution of the adjacent area of the predictive point could be more fully considered,and obtain the HDR image approximately to the natural scene.In addition,this thesis also uses a variety of inverse tone mapping operator to directly stretch the dynamic range of LDR images,and a new stretching method with good imaging results is proposed.The experimental part of this thesis compares a variety of HDR imaging methods,and the subjective and objective aspects of the quality of the resulting images were evaluated.The experimental results show that the inverse CRF estimated by the proposed algorithm has better normality,consistency,and stability.In the effect of HDR imaging,this algorithm can retain more complete details in highlight and shadow areas without inducing contrast distortion,and present more natural brightness and chrominance information.The tone mapped version of the HDR image is less different from the original LDR image,and it can reflect the realistic scene.This method can meet the practical application requirements,and is a good choice for HDR imaging.