Study On Methodology And Technology Of Acquisition And Rendering For High Dynamic Range Color Images

Currently, there is a trend toward higher spatial resolution as well as higher dynamic range in imaging technology. However, the development of the imaging hardware lags behind human demand for acquisition and rendering of HDR (high dynamic range) image, e.g. the dynamic range of digital cameras is usually lower than the dynamic range of HDR scenes, and the mainstream displays cannot reproduce the HDR scenes accurately. This study investigates basic theoretical methodology and key practical techniques comprehensively, focusing on the accurate color communication during the processing of HDR image acquisition and rendering. Firstly, it was discussed on how to capture the luminance and chrominance of HDR scenes accurately with commercial digital cameras, followed by restoring them in HDR image formats. In order to render HDR images in LDR (low dynamic range) displays, the tone mapping/reproduction operators, which compress the HDR images into LDR images, were discussed systematically. It would be helpful to improve the colorimetric accuracy of HDR image acquisition if the SPD (spectral power distribution) of the HDR scene illumination was obtained. Hence, SPD reconstruction techniques of light sources were introduced and then the recovery methods for fluorescent lamps and multi-channel LED light source were proposed respectively. Finally, an application instance, which combines the main methods in acquisition and rendering of HDR color image, was implemented to verify the achievements of this study.Software-level approach by fusing multiple images of different exposure was adopted in this study for the purpose of capturing HDR scene using commercial digital imaging devices. Compared with traditional fusing technology, this study focused on the colorimetric information of HDR image capture. Derived from the imaging pipeline of digital cameras, two different colorimetric characterization models of digital cameras considering capture settings were proposed, i.e. one based on homogeneous polynomial regression, the other based on searching and scaling. In different lighting conditions provided by two kinds of light booths, three different levels of digital cameras with different brands were employed to test and verify the effectiveness of proposed methods. And the performance of the proposed methods influenced by different capture settings as well as their different parameters was investigated. Then based on the colorimetric characterization of digital camera with unrestricted capture settings, an HDR image fusing technique in device independent color space, such as CIEXYZ color space, was proposed and then demonstrated to be effective.Because the bit depth of the HDR image is beyond that of the mainstream LDR displays, HDR images cannot be displayed directly and need to be compressed into LDR images. By comparing several typical colorimetric characterization models of displays, GOG (gain-offset-gamma) model was chosen to calibrate the EIZO display which was used in the experiment. Several typical tone mapping operators were compared via psychophysical experiment. As for the iCAM06 (image color appearance model) that outperforms the others, it was discussed the optimization for the parameter of max luminance in iCAM06. Because the hue-shift problem and the complicated computation process exists in iCAM06, a modular tone mapping algorithm based on iCAM framework was proposed and proved to be valid by a paired comparison experiment with iCAM06.The colorimetric characterization model of digital cameras in HDR image capture still has trouble in generalization for light sources, i.e. the model trained under one kind of light source may not be applicable for the prediction under another kind of light source, especially when the two kind of light sources differ greatly. As a result, the reconstruction technique by imaging was considered. Several reconstruction methods for object reflectance in common use were introduced and transformed to the recovery techniques for light source SPD. Simply employing RGB camera to capture light source and recover its SPD may cause metamerism problem, therefore the standard color chart with known reflectance was used as a’supervisor’. Capturing the different color patches is a mathematically equivalent way of recording the light source by the multispectral technique. Fluorescent lamps are the most widely used sources, and a modified algorithm based on classification and channel optimization was proposed. Spectrally tunable light source system of multi-channel LEDs is the most popular source nowadays, an algorithm based on matrix factorization was proposed to recover its SPD.In order to further investigate the validity of the research achievements in this study, a workflow combining the techniques of HDR image acquisition, HDR image rendering and light source SPD recovery was proposed. An instance of capturing and rendering road lighting at some place on yuquan campus of Zhejiang University was carried out as the HDR imaging application. HDR image was firstly fused from multiple images with different exposures based on colorimetric characterization model of digital camera, and then rendered by the proposed tone mapping algorithm based on iCAM framework, the SPD of road lighting was recovered as well. The instance demonstrates the effectiveness and practical applicability of the methodology in HDR imaging. Besides, two main factors which influence the luminance measurement by imaging in practical use were analyzed, i.e. the non-uniformity of imaging system (or shading) and the focus distance impacting on output values, and the correction coefficients were calculated accordingly.Finally, the main contents and innovations of this dissertation were summarized, and the perspectives of the future study were also forecasted.