Abstract
High dynamic range (HDR) images require tone-mapping to scale the large range of luminance information that exists in the real world so that it can be displayed on a device that is capable of only a limited dynamic range in luminance, to adequately reproduce their perceptual qualities. Here the main objective is to preserve the original HDR scene appearance, including contrast, sharpness, and color with the focus on overcoming the limitation of the output media to achieve the best match given the limited gamut and dynamic range. To the best of our knowledge, these HDR rendering algorithms have been proposed only for traditional three-channel (RGB) images. It is well known that having multispectral data can produce better color-accurate images, but whether the same thing can be said for tone mapping too or not, has not been explored yet. In addition, limited research work regarding tone mapping for HDR hyperspectral images is itself a challenge whereas there is no such publicly available database for HDR hyperspectral images.
To address these issues, this master’s thesis aims to investigate the effective utilization of spectral radiance for improved color fidelity and tone-accurate reproduction of HDR images. To achieve this, an HDR hyperspectral radiance image has been captured using an approach similar to the multiple exposures HDR technique which helps to significantly recover the details of a high dynamic range (HDR) scene both in the dark and bright areas, overcoming the problem of capturing underexposed and overexposed data. The HDR hyperspectral absolute radiance image is further improved by applying the proposed linearity correction method while hyperspectral interpolation has been performed to verify and account for the missing wavelengths due to pixel saturation. It also introduces a spectral image color appearance model titled SiCAM which is the first of its kind, designed for tone mapping a HDR hyperspectral radiance image to a three-channel Low Dynamic Range (LDR) image. It is to be noted that SiCAM is inspired by the iCAM06 image color appearance model where we adapt the iCAM06 for hyperspectral input by embedding a spectral adaptation method rather than chromatic adaptation as in the case of iCAM06 which requires only three-channel RGB input.
Additionally, a psychophysical experiment has been conducted for perceptual evaluation of the proposed HDR rendering method to determine the effectiveness of having more spectral data for the tone mapping of HDR images in comparison to the performance of iCAM06 and the gamma operator. Besides, the objective image quality assessment (IQA) of these reproduced LDR images has been performed. The results from both subjective and objective evaluation indicate that SiCAM outperformed the other two HDR rendering methods in terms of both accurate color appearance and pleasantness. Finally, we presented a dataset containing four HDR hyperspectral radiance cubes and their respective three-channel HDR images as our contribution to future research in this domain.