A Brief Research On Interpolation And Denoising Of CFA Images

A typical consumer digital camera uses one single-chip CCD or CMOS sensor with a Color Filter Array (CFA) to capture images which are named as CFA images. Only one color component is sensed at each pixel in a CFA image, with the other two missing. The full color image is reconstructed by interpolating the missing color components, the process known as demosaicing. In order to obtain high-quality images, it is nessisary to suppress the noise introduced in the imaging procedures. The interpolation and denoising of CFA images are the key stages in the procesing of digital images. The research on CFA images are of great commercial and academic value.In this paper, an introduction to the basis of CFA image processing is given. Many existing demosaicing algorithms are overviewed and two frequently-used measurements are introduced. As for Bayer pattern CFA images, this paper propose two demosaicing methods and a CFA adaptive denoising method. The major works can be summarized as follows:(1)A new edge-directed demosaicing method based on luminace estimation is presented. This approach estimates the luminance of CFA images to guide the interpolation of G channel, and recovers R and B channels with the help of bilinear interpolation on the color differences. The performance is improved with a refining process in which the edges of the final results are enhanced. The effectiveness of proposed method is verified by the experiments on Kodak image database compared with several state-of-the-art methods.(2)A demosaicing method via directional interpolation and residual image reconstruction is proposed. The residual image between the initial demosaicing G channel and the ground-truth is reconstructed with two redundant dictionaries jointly trained by pre-interpolated G component of a training image set. A scaled version of the residual image is added back to the initial interpolated G channel. Thus, a better estimate of the G channel is obtained, which contributes the improvement of the final results.(3)A spatial adaptive denoising method based on Treelet is proposed for noisy CFA images. This approach extracts training data from the high-frequency part of CFA images, and transforms the data to Treelet domain. With the shrinkage of the coefficiences, the noise of CFA images is adaptively suppressed. The effectiveness of proposed method is verified by the experiments compared with a PCA-base denoising method and a joint denoising and demosaicing method.