Higher Order Fitting based Image De-noising using ANN

The images usually contain different types of noises while processing, coding etc. Image usually contains noise because of poor transmission purposes. As a result, it produces bad image which is difficult for processing purposes. The currently available methods are usually based on wavelet transformation. Some of these methods still have problem in tackling with noise. This paper presents an artificial neural network approach to de-noise an image even if the level of noise is high .The training algorithm involves scale gradient conjugate back-propagation. In this paper, we have worked with salt and pepper noise as well as with Gaussian noise. We have proposed a method in which the noise will be detected by the algorithm and the neural network will be trained accordingly. After that a Feature Vector Table (FVT) will be prepared. On the basis of FVT neural network will be trained. Experimental result shows that our proposed method could achieve a higher peak signal-to-noise ratio (PSNR) on images as compared to other methods. Keywords—Artificial neural network, training algorithm, image de-noising, salt and pepper noise, Gaussian noise