Using neural networks for undithering

The dithering is one of the techniques for transforming a 24-bit image to an 8-bit image. It has an advantage over other transformation techniques in that the information of 24-bit colour values is still accessible in a form of dithered patterns of an 8-bit dithered image. A transformation of an 8-bit dithered image back to a 24-bit image (undithering) could be done by estimating 24-bit colour values from the 8-bit dithered patterns. This study applied a neural network for undithering. The backpropagation method was used for training the neural network. The fully trained neural network was used for undithering the testing images. The evaluations were done by comparing the original 24-bit images and the undithered 24-bit images in two aspects. Firstly, the statistical comparison was done by using a histogram. Secondly, from an aesthetic point of view, the 24-bit undithered images were compared to the 24-bit dithered images. The results shown that, the undithered images are not as sharp as the original image, but, a large number of colours were reproduced compared to the dithered image. An important factor that affects the performance of the neural network is the characteristics and range of colours of the images used for training the neural network