Color Filter Array Demosaicing by Using Wavelet Transform Scheme

Digital still cameras use image sensors to capture images. To reduce the size of digital still cameras, a single image sensor instead of three image sensors should be used. A single image sensor overlaid with a color filter array (CFA) that only captures one color channel at each pixel; thus, the other two missing color channels at each pixel must be reproduced from neighboring pixels. This procedure is known as CFA interpolation or demosaicking. The main goal of this thesis is to design and analyze algorithms of CFA interpolation. Our research focuses on two CFA interpolation systems: classifiers for edge detection and demosaicking by using subband synthesis. At the first part of this thesis, we introduce the background of CFA interpolation and the research motivation. Then, the traditional techniques of CFA interpolation are reviewed. They can be categorized into three groups which are non-edge-sensing algorithms, edge-sensing algorithms and frequency correlation algorithms. Classifiers for edge detection are important in CFA interpolation, since interpolating a missing pixel along an edge or across an edge can result in apparently different interpolation. In the second part of this thesis, classifiers for edge detection are introduced and a novel wavelet-based interpolation algorithm is presented to predict the direction of edges. In CFA interpolation process, there are two artefacts, i.e. zipper effect and false color artefacts may affect the CFA interpolation performance. In the third part of this thesis, a wavelet based CFA interpolation method is presented to reduce zipper effect and false color artefacts. First, an edge detection method is used to interpolate the initial missing color channels. Second, the high correlations between wavelet subbands of the different color channels are explored to obtain accurate green color values of the estimated green channel. Finally, the high-frequency subbands of red and blue channels are iteratively updated to reduce false color artefacts. In this dissertation, effective classifiers for edge detection and a wavelet based CFA interpolation algorithm are presented. Experimental results demonstrate that our proposed method can indeed preserve edges and restore images with high quality.