| This thesis presents a systematic study of high dynamic range (HDR) imaging that considers the entire imaging processing system, including properties of natural scenes, imaging sensors, image processing algorithms, image displays and image perception.; We begin by studying the statistical properties of more than 70 captured HDR images of natural scenes using a customized acquisition device. The dynamic range of many of these recorded images is higher than 10000:1. However, if we exclude the one percent brightest pixels (usually created by specular reflections or light sources) and the one percent of the darkest pixels (usually created by dark shadows), the dynamic range of most images is reduced to less than 1000:1.; In order to characterize the human visual system's response to HDR images, it was necessary to build a HDR display that is capable of displaying the whole range of these images. Inspired by the surface-illumination composition of HDR scenes, we built a display device that combined a DLP projector and a raw LCD panel to display image data with intensities ranging from 1 cd/m 2 to 10,000 cd/m2.; The HDR device was used to study the effect that color, background luminance levels and surround luminance levels have upon human visual sensitivity to noise. The results show that noise is barely visible when the luminance noise contrast level (defined as the ratio of the standard deviation in luminance noise over the mean background luminance) is less than three percent.; Based on these results, we showed that an ideal capture device with only photon-shot noise must be designed to capture at least a thousand photons per pixel (or 30 dB SNR) across the whole image to avoid visible noise. To encode the dynamic range of most natural images without visible noise, traditional single capture devices need a full well capacity of at least 106 while multiple capture devices need a much smaller full well capacity that is independent of scene dynamic range. We also used the perceptual noise threshold results to design a new multiple capture exposure scheme. |
