ANALYSIS OF DECONVOLUTION ALGORITHMS FOR IMAGE RESTORATION

Abstract

This report describes the latest developments in the field of image deblurring.The methods of image restoration try to remove the noise and blurring while capturing the image .The concept of deblurring is to generate a better representation of the original blurred image. Combination of the blurring and random noises make image deblurring process very problematic.To create the well-posed solution we need to merge additional information about the ideal image. From several decades significant progress is continuously being made in image restoration field. The applications of image restoration are in the field of medical imaging, space exploration, stegnography etc. In this study analysis of various blind and non blind methods is done for restoring the picture .Various non blind methods like inverse filtering, pseudo inverse filtering and Weiner filtering ,Lucy Richardson(LR) are studied. All these methods are non blind methods so they have knowledge of PSF .But in case of absence of known PSF these methods would not work .These methods have drawbacks like weiner filtering method shows slow convergence rate, LR method shows ringing effect and high processing time and inverse filter is very sensitive to noise .But as we studied Blind methods then we analyzed that blind methods are giving better results as compared to non blind methods. Blind methods do not contain the information regarding PSF. So there is a need to estimate the PSF. Advantage of blind methods is that they improve the convergence rate of the process to get deblurred image. These methods remove the ringing effects from the restored image and take less processing time and also preserve ii

the edges to restore the better results in terms of image intensity around the edges. The applications of this method exist in the field of aerospace and defence sector, also in field of medicine and robotics. In this study we analyzed various methods and their simulation and experimental results were compared with each other.As a result we observed that deblurring using blind deconvolution is getting better results as compared to other methods in terms of better convergence rate and less mean square value.