하이 다이나믹 레인지 영상의 디스플레이를 위한 효율적인 톤 재생 기법

Abstract

The ultimate purpose of the engineers who work for image processing and vision is to represent a scene that can be achieved the perceptual match between real scene and reproduced scene by digital devices. To achieve this goal, a variety of the techniques has been developed to mimic the human eye and vision. High dynamic range imaging (HDRI) is a promising technique to solve this problem that can describe the range of light intensities in the natural world and even can record much broader range of luminance than the human eye can recognize. Examples of applications for HDR imaging include scientific and medical visualization, satellite imagery, digital camera, and digital cinema. In particular, the ability of capturing the real world luminance in a scene has become a necessary function for a digital camera. However, compared with the 108:1 range of real world luminance from bright sunlight to starlight for HDRI, display devices such as plasma/CRT/LCD displays and projectors have low dynamic range (LDR) near 102:1 cd/m2. Although high dynamic range (HDR) monitors will be more widely available in the near future, they are still rare and costly and, at the same time, difficult to calibrate. Therefore, to present HDR images for displaying on conventional display devices, the range of values must be transformed to a displayable range that covers no more than two orders of magnitude, a process called tone reproduction or tone mapping. The tone reproduction is an essential processing to convert an HDR image into a LDR image, but the details of extremely dim and bright regions may be lost in common LDR images when simple tone reproduction is applied. Therefore, in order to guarantee a perceptual match between the real scene and tone reproduced image in conventional display devices, proper tone reproduction techniques, which not only transform HDR images into displayable range, but also preserve details, are required by considering the characteristics of the original HDR image. Tone reproduction can be classified as a global tone, in which the same reproduction function is applied in all regions, or a local reproduction, in which different tone reproduction functions are applied throughout the modeling of spatial adaptations, depending on neighboring pixels. This dissertation proposed a global operator (piecewise linear function-based method) and two local operators (chromatic adaptation-based method and filtering-based method) without causing the drawbacks in each method. The proposed techniques basically focus on enhancing visual fidelity of tone reproduced image through the use of the concept of human visual system (HVS) and color appearance model (CAM). The piecewise linear function-based method is a global tone reproduction algorithm, but the algorithm achieves good perceptual performance. The proposed piecewise linear reproduction method could overcome a common drawback of global tone reproduction methods, which cannot successfully preserve the local edge contrasts of the image especially in the dim and bright regions. The chromatic adaptation-based method reproduces an HDR image on block basis to apply a color appearance model adaptively. Instead of relying only on reduction of dynamic range, the chromatic adaption technique is adaptively applied to preserves chromatic appearance and color constancy across scene and displayed scene. Lastly, the main goal of the filtering-based method is to enhance contrast while reproducing an HDR image. The algorithm mainly consists of a histogram-based luminance adjustment and an adaptive Gaussian filter using minimum description length (MDL). After applying a reduction of the dynamic range according to a histogram-based luminance adjustment process, and then the contrast is enhanced based on the Gaussian filtered image with an optimal variance. The main contribution of the chromatic adaptation-based method and the filtering-based method are local operators, but they reproduce an HDR image without causing the common drawbacks of local tone reproduction methods: high computational complexity and undesirable artifacts. The proposed algorithms have been tested on several sequences and found to provide better visual quality than conventional tone production schemes, especially in dark and bright region. In particular, the proposed schemes have fast, simple and practical structure for implementation.