Towards Joint Super-Resolution and High Dynamic Range Image Reconstruction
The main objective for digital image- and video camera systems is to reproduce a real-world scene in such a way that a high visual quality is obtained. A crucial aspect in this regard is, naturally, the quality of the hardware components of the camera device. There are, however, always some undesired limitations imposed by the sensor of the camera. To begin with, the dynamic range of light intensities that the sensor can capture in its nonsaturated region is much smaller than the dynamic range of most common daylight scenes. Secondly, the achievable spatial resolution of the camera is limited, especially for video capture with a high frame rate. Signal processing software algorithms can be used that fuse the information from a sequence of images into one enhanced image. Thus, the dynamic range limitation can be overcome, and the spatial resolution can be improved.
This thesis discusses different methods that utilize data from a set of multiple images, that exhibits photometric diversity, spatial diversity, or both. For the case where the images are differently exposed, photometric alignment is performed prior to reconstructing an image of a higher dynamic range. For the case where there is spatial diversity, a Super-Resolution reconstruction method is applied, in which an inverse problem is formulated and solved to obtain a high resolution reconstruction result. For either case, as well as for the optimistic and promising combination of the two methods, the problem formulation should consider how the scene information is perceived by humans. Incorporating the properties of the human vision system in novel mathematical formulations for joint high dynamic range and high resolution image reconstruction is the main contribution of the thesis, in particular of the published papers that are included. The potential usefulness of high dynamic range image reconstruction on the one hand, and Super-Resolution image reconstruction on the other, are demonstrated. Finally, the combination of the two is discussed and results from simulations are given.
Human visual system
Digital camera system
HC3, Hörsalsvägen 14, Chalmers University of Technology
Opponent: Professor Fredrik Kahl, Department of Signals and Systems, Chalmers University of Technology, Sweden