A parallel approach to train FLANN for an adaptive filter

Images get corrupted at the time of transmission due to various noises, such as additive white Gaussian noise, impulse noise or both of these two, salt and pepper noise, multiplicative noises, random value impulse noise and many more. Neural network based image filter is one of the most important example of adaptive image filter. Adaptive neural network filter remove various types of noise such as Gaussian noise and impulsive noise. Neural networks have already been applied in several domains of image processing including image filtering. But training of those neural networks consume much time before it is actually tested on such as image filtering. Applying parallelism to image processing is increasingly practical and necessary, as our desktops are becoming multi- core machines replacing single core. Therefore, this paper presents a parallel approach called image decomposition technique to train FLANN (Functional Link Artificial Neural Network) before it is actually used for rectifying the corrupted pixels to restore the image. Experimental results obtained through SPMD (Single Program Multiple Data) simulation environment show that the proposed parallel approach to train the FLANN is feasible as it substantially reduces the training period and also make it an efficient filter to restore the image fairly well maintaining the quality of the filtered image.