Abstract:
Modern digital still cameras are equipped with just a single CCD for color image acquisition. Since only one spectral band can be recorded in each pixel, a mosaic of red, green and blue color filters is placed in front of the chip. The process of subsequently calculating a full color image from the partial sampling is referred to as demosaicking. Simple linear interpolation leads to blurring and noticeable color artifacts in the reconstruction image. This thesis discusses the application of methods from spatial supervised learning to the demosaicking problem. A statistical model based on conditional Gaussian Markov random fields (GMRFs) is derived and suitable features are selected. The conditional shows competitive results to existing methods with respect to numerical error measures as well as by visual assessment.
