Blind Image Restoration Algorithm

Abstract

The objective of Image Restoration is to reconstruct the original image from a de- graded observed image. This recovery process is critical to many image processing applications. The classical restoration algorithms generally assume that prior knowledge of the nature of the degradation and noise contained in an observed image is avail- able. When the image is restored from the degraded image only, without any prior knowledge of degradation occurred, then the process is called Blind Image Restoration.

Basic terminology and existing methods for Image Restoration are studied as part of the thesis. Degradation of an image and the effects of different degradation functions such as Motion, Gaussian and Defocus blur on different images are studied theoretically and experimentally. A Graphical User Interface (GUI) is developed for the better understanding of the image restoration process. In GUI applied original image can de- graded using three blurring functions, Motion Blur, Gaussian Blur and Defocus blur with various parameters and then the restoration is done using filters such as Inverse filter, Wiener filter, Regularized filter and Lucy-Richardson filter.

There are a plethora of methods present for restoration of images with specific kind of blurs but a generalized solution for any kind of blurred image restoration is not available. A novel Blind Image Restoration algorithm based on Moment Invariants is proposed as a general solution to restore the degraded images due to mostly occurring three kinds of blur i.e. Motion blur, Gaussian Blur and Defocus Blur. The proposed algorithm first detects the type of blur present in the degraded image and then generates Point Spread Function (PSF) by estimating the blur parameters for the detected blur. Restoration is done using the classical restoration filters such as Wiener, Regularized and Lucy-Richardson filter. Experimental results show that the proposed algorithm is less complex, requires less computational time and provide accurate results with high efficiency.