Color Image Reconstruction for Digital Cameras
Daniele Menon · Padua Research Archive (University of Padova) · 2009
Recently we observed a fast diffusion of digital cameras that are able to acquire images and videos directly in the digital format. This new acquisition technique allowed to explore new strategies to process, save and display images and videos. Digital cameras require many operations to process the data acquired by the sensor. In this thesis I present an overview of the techniques used in practical realizations and proposed in the literature. Particular attention is paid to the algorithms that are more connected to the image processing area. Among them, the process that is the most important for the quality of the resulting images and the most computational demanding is demosaicking. This consists in the reconstruction of the full color representation of an image from the data acquired by a sensor provided with a color filter array that in each pixel acquires a color component only instead of the three values that are necessary to represent a color image. The most common color filter array is called Bayer pattern, from the name of his inventor Bryce Bayer. In this thesis an overview of the demosaicking techniques presented in the literature is given and three new methods that allow to obtain good performances with a reduced computational cost are proposed. The first two are based on directional interpolations and are made adaptive to the image behavior through analysis of the edges and wavelet transformations. The last proposed technique, instead, is based on regularization methods, an useful tool to find a solution for an ill-conditioned inverse problem. Since the sensor introduces a noisy component in the acquired data, an algorithm to perform demosaicking and denoising jointly is also analyzed. It exploits wavelet transformations. Finally, a method to adaptively interpolate the image is presented, in order to increase the resolution and improve the visual quality of the details in the image. This technique is based on an analysis of the statistical local behavior of the image.