ABSTRACT:In image deblurring mathematical models are used to seek or recover the original image. The key issue is that some information assumed to be lost is present in the blurred image. But this information is "hidden" and can only be recovered if the blurring process detail's is known. In this paper the advantage of adaptive filter theory with neural network is combined to create a new deblurring model. The proposed deblurring model deblurr the images with significant improvements. The neural network present in the proposed model is first trained with a set of blurred and original image pair's and once the neural network is trained it is used to deblurr similar kind of blurred images. KEYWORDS: Deblurring, Artificial neural network, Functional link artificial neural network (FLANN), multilayer perceptron (MLP), and Point spread function (PSF).
INTRODUCTION:Over the last several years, Image Deblurring has been an active area of research and it has been considered for a variety of applications in signal and Image Processing.Deblurring is an active area research for a long time. This field comes into existence with the advancement of digital photography. When we capture a moving object with a slow capture speed camera or vice versa the image becomes blurred. We need to remove this noise from the image. With the requirements of aerial surveying, digital photography, it has tremendous application.Recently, artificial neural networks (ANN) have emerged as a powerful learning technique to perform complex tasks in highly nonlinear dynamic environments. Some of the prime advantages of using ANN models are: their ability to learn based on the optimization of an appropriate error function and their excellent performance for approximation of nonlinear function. [1] The functional link ANN (FLANN) proposed by Pao [2] has shown that this network can be used for function approximation and pattern classification with a faster convergence rate and lesser computational complexity than an MLP network.The performance of the FLANN for the task of identification of nonlinear systems has been reported. [3] Using trigonometric functions as functional expansion, the superior performance of the FLANN with respect to the MLP network has been obtained. Here, we propose an alternate FLANN structure, which has been shown to provide effective identification of nonlinear dynamic systems. For functional expansion of the input pattern, Chebyschev polynomials are used in [5][6] instead of trigonometric functions. Being a single layer neural network, its computational complexity is less intense as compared to (MLP) and can be used for online learning. Pattern classification using CNN has been reported in. [6] Blurring makes an image indistinct and hazy in outline or appearance. Blurring caused by motion, hazy and out of focus lens of the camera.