Colorization of gray scale natural still images by using ANN to predict the low frequency DCT components of the RGB channels

This paper presents a new algorithm for colorizing gray scale natural still images. The algorithm uses artificial neural network (ANN) to predict the low frequency discrete cosine transform (DCT) components of the RGB channels. A set of natural color images are used to train three ANNs. The trained networks estimates the RGB layers of the gray scale image that best match a set of training colored images. The ANN predicts only the low frequency components. The high frequency components of the gray scale image are mapped to the RGB channels. The performances of the new algorithm are analyzed using the peak signal to noise. Acceptable colors were obtained for a variety of still images.