A learning-based approach to image demosaicking with spatial autocorrelation analysis

초록

We introduce a two stage image demosaicking method for Bayer color filter array (CFA) images. Pixel interpolation using a Bayesian and/or SVM classifier is followed by renegotiation of the interpolated image with an auto-correlation function (ACF), which is applied to the distribution of edge strengths at each pixel of the interpolated image. This second stage can also be used to post-process images produced by other demosaicking methods. Experimental results obtained with the Kodak PhotoCD benchmark show that our method shows enhanced edge and texture details and when compared with three other methods.