Feasibility of less complex Viterbi decoder based on neural networks for effective transmission of medical images

Abstract

Convolution coding is the most commonly used channel encoding technique for reliable data transmission. At the receiver's end, Viterbi decoder is used for extracting the message bits. Though these decoders can correct any number of errors, complexity of the decoder increases with the number of bit errors. Hence research began in this area, to design decoders which are less complex and yet provides accurate results. In this paper, a five layered feed forward neural network is developed to decode the image pixels which are convolutionaly encoded. The decoded intensities are then converted into an image and the edges are detected from the images. Edges are regarded as useful information because segmentation of the Region of Interest (ROI) is performed on the edge detected image. Performance of the proposed technique is measured in terms of Root Mean Square Error (RMSE) and Peak Signal to Noise Ratio (PSNR). It is found that irrespective of the images, RMSE is significantly less and PSNR is high and the true edges are retained in the image. The proposed technique is scalable as it is not dependent on the size of the image but on the gray levels of the image.