Demosaicing using Artificial Neural Networks

The problem of color image enhancement and the specific case of color demosaicing which involves reconstruction of color images from sampled images, is an under-constrained problem. Using single-channel restoration techniques on each color-channel separately results in poorly reconstructed images. It has been shown that better results can be obtained by considering the cross-channel correlation. In this paper, a novel approach to demosaicing is presented, using learning schemes based on Artificial Neural Networks. Thus the reconstruction parameters are determined specifically for predefined classes of images. This approach improves results for images of the learned class, since the variability of inputs is constrained (within the image class) and the parameters are robust due to the learning process. Three reconstruction methods are presented in this work. Additionally, a selection method is introduced, which combines several reconstruction methods and applies the best method for each input. The first reconstruction method presented in this work is the Perceptron method which considers linear restoration models. Reconstruction using the Perceptron method yields excellent results in low-frequency regions of the image. However, in the high-frequency regions (especially in desaturated color regions) the results of the Perceptron network are not satisfactory. This is shown to be due to the linearity and the uniformity of the Perceptron network. Restoration using the Backpropagation method yields relatively good results in high-frequency regions. However, in the low-frequency regions the results are not satisfactory since it fails to reconstruct the correct color of the region. The Selector method is shown to be useful in combining both of the previous methods in one reconstruction scheme which works in both cases of low and high frequency regions. The Quadratic Perceptron method is a new development extending the Perceptron network in order to overcome the limitations of the latter network and provide a better solution for the reconstruction in the high-frequency regions.