Development of blind deconvolution technique for restoration of gray scale image

Abstract

Image restoration schemes are based on PSF estimation. Based on various aspects of restoration, three different contributions have been made in this thesis. This approach is used for the removal of blur using MAP based image restoration technique. The comparative analysis of the cepstrum based optimization techniques has also been done. In this research work, implementation, analysis and identifying the correctness of existing methods namely Radon transform, Hough transform, MAP based technique, Patch-based technique to detect PSF have been done. But the major issues faced with the existing PSF estimation method are less accuracy and restriction on the types of blur handled by these methods and high computational cost. Three new methods to overcome the above mentioned problems namely: An improved MAP based approach of BID, optimized modified Cepstrum based blind deconvolution technique and Deep learning-based blind deconvolution technique of image restoration have been proposed, designed and implemented. After comparative analysis, it has been observed that these approaches have better accuracy, flexibility and low computational cost as these are scalable, which improves the PSF estimation of the blind image deconvolution technique of image restoration. Standard grayscale images, the standard database of greyscale images and the GOPRO database are used. It has been observed that improved MAP based technique increased accuracy, but it generated some ringing effects in the restored images. It has also been observed that the improved cepstrum based method of BID of image restoration works only for motion blur type. This limitation is overcome by the Deep learning-based blind deconvolution technique of image restoration. These methods also show noticeable improvements in the time required for PSF estimation techniques. newline