Image denoising by hybridizing preprocessed discrete wavelet transformation and recurrent neural networks

Abstract

In this paper, we are analyzing the performance of Recurrent Neural Network (RNN) for image recalling with improved training sets by Discrete Wavelet Transformation (DWT). DWT has been used for decomposing the images into four parts for low level feature extraction, to build the pattern information, encoding the pattern. When all the patterns of these four level training sets are encoded, and given as input to RNN to analyze the performance. This analysis is carried out in terms of successful and correct recalling of the images by hybridizing of DWT and RNN. Now we introduce salt and pepper noise so that the distorted feature vectors presented to the network. This gives a prototype pattern of noisy image and requires filtering of the training set. This leads to recalled output of the network that produces the pattern information for each part of the images. Now the integration is made possible if inverse discrete wavelet transformation (IDWT) to amalgamate the recalled outputs corresponding to each part of the image and final image is recognized.