Academic Journals Database
Disseminating quality controlled scientific knowledge


Author(s): Ananya Pothula | Ch.Navya Deepthi | P. Suresh

Journal: International Journal of Computer & Electronics Research
ISSN 2320-9348

Volume: 2;
Issue: 3;
Start page: 206;
Date: 2013;
Original page

Keywords: Neural networks | multi-layer perceptron | simple multilayer perceptron | Graphics processing units

The technique of image denoising proceeds a noisy image as input and yields an image where the noise has been condensed. Neural networks have already been used to denoise images.  It is probable to attain state-of-the-art denoising presentation with a simple multilayer perceptron. Multilayer perceptrons can be assumed of as widespread function approximators and are more commanding. A multi-layer perceptron is a nonlinear function that maps vector-valued input by means of numerous hidden layers to vector-valued output. The design of a multi-layer perceptron is defined by the number of hidden layers and by the layer sizes and it is similarly named as feed-forward neural network. The parameters of the multilayer perceptrons are projected by preparation on pairs of noisy and clean image patches by means of stochastic gradient descent. Multilayer perceptron maps noisy patches onto noise-free ones which is probable due to the capability of the multilayer perceptron selected is vast sufficient containing of adequate hidden layers with sufficiently numerous concealed elements. The patch magnitude is selected great sufficient containing sufficient information to improve a noise-free version. Training illustrations are produced on the fly by humiliating noise- free patches with noise. Multilayer perceptron is personalized to a single level of noise and does not simplify well to other noise levels linked to other denoising methods. Working out high capacity multilayer perceptron with big training sets is possible by means of modern Graphics Processing Units.
Why do you need a reservation system?      Save time & money - Smart Internet Solutions