Image Quality Enhancement Using Domain-specific Priors

Abstract

The development of sophisticated imaging technologies has made cameras being not only widely used in our daily lives, but also capable of imitating the functioning of the human eye. Despite the advances of imaging devices, however, there is still room for improving the quality of obtained images because imaging systems are not perfect and can introduce some amounts of distortion or artifacts in the signal at any time. As high quality images are frequently required for various graphics and vision applications, image quality enhancement to overcome the fundamental limits of photography is crucial and necessary. Image quality enhancement aims to improve the interpretability or perception of information in images for human viewers and provide better input for other image processing techniques. These days, as digital cameras have come into wide use, the effort of improving the image quality is mostly accomplished by computer software for digitally stored images. Accommodating their popularity, this thesis presents software-based image quality enhancement solutions for digital photographs. Recent studies for image quality enhancement mostly involve the development of image priors on top of a well-defined image restoration framework. Since image priors capture common properties of concerned images and help to cope with the ill-posed nature of the problem, the image prior knowledge plays a critical role in image quality enhancement. However, despite the success of image priors, general-purpose image priors are too weak to regularize a particular problem domain and often do not provide satisfactory results. To overcome the limitations, domain-specific priors that account for the characteristics of a problem domain have been getting more attention recently. As domain-specific priors explicitly consider domain knowledge, it is considered as the key to produce high quality results. In this thesis, we follow the trend that exploits domain-specific image priors for image quality enhancement. Specifically, this thesis includes the following topics: Text Image Deblurring Using Text-specific Properties - Previous blind image deconvolution approaches have difficulties when dealing with text images, since they rely on natural image statistics which do not respect the special properties of text images. In this work, we propose a novel text image deblurring method which takes into account the specific properties of text images as a domain-specific prior. Our method extends the commonly used optimization framework for image deblurring to allow domain-specific properties to be incorporated in the optimization process. High Dynamic Range Imaging Using Coded Electronic Shutter - Typical high dynamic range (HDR) imaging approaches based on multiple images have difficulties in handling moving objects and camera shakes, suffering from the ghosting effect and the loss of sharpness in the output HDR image. In this work, we propose an HDR imaging approach using the coded electronic shutter which can capture a scene with row-wise varying exposures in a single image. We specifically investigate the characteristics of the coded electronic shutter as domain-specific prior knowledge and utilize it to produce a desired HDR image without ghosting and blur artifacts. Vignetting Correction Using Radial Bright Channel Prior - Previous vignetting correction approaches require reference calibration images or multiple images of different intensity attenuation with known camera settings. In this work, we presents a radial bright channel prior for single image vignetting correction, derived from a statistical property of vignetting-free images. Exploiting the prior, we can effectively estimate and correct the vignetting effect of a given image.