Enhanced Medical Image De-noising Using Auto Encoders and MLP

Abstract

Preserving the original characteristics of an image which is transmitted across a channel having different kinds of noise (i.e., either, uniform, linear or Gaussian noise) is a crucial task, hence it has become a state of art for the researchers in retrieving the original characteristics of the image by using different denoising and image retrieving techniques. In earlier, many techniques have been proposed such as patch wise denoising (e.g., Sliding Window), block matching (e.g., BM3D), shallow and wide deep learning algorithms which achieved a promising accuracy, yet failing in preserving the prominent characteristics of an image which is a crucial task in Bio-Medical Instrumentation systems. So, we proposed few algorithms which could preserve the smallest possibilities of denoising the medical images and achieved a maximum accuracy of 99.98% for SDAE (In Tensorflow Background), 99.97% for SDAE (In Theano Background) and 99.99% for Multi-Layer Perception (MLP) technique and later compared these with the accuracies of the existing methods.