© 2014, IJARCSMS All Rights Reserved 280 | P a g e ISSN: 232 7782 (Online) 1 Computer Science and Management Studies  International Journal of Advance Research in Volume 2, Issue 9, September 2014 Research Article / Survey Paper / Case Study Available online at: www.ijarcsms.com Image and Video Deblurring Algorithm Using Normalized Sparsity Measure Rosy John 1 Department of Electronics and Communication Engineering Amal Jyothi College of Engineering Kanjirappally – India Ajai Mathew 2 Department of Electronics and Communication Engineering Amal Jyothi College of Engineering Kanjirappally – India Abstract: Motion blur is an artifact that causes visually annoying images. In this work a novel algorithm to deblur blurred images and video frames is proposed. Prior to deblurring it automatically checks whether an image or frame is blurred or not using a method based on Cumulative Probability of Blur detection. If the image is blurred, deblurring algorithm is applied obtain the true image. This method utilizes normalized sparsity measure to recover the sharp image. The algorithm is simple fast and robust. Experimental results show that this method can effectively remove motion blur and recover sharp image. Keywords: motion blur; deblurring; cumulative probability; normalized sparsity measure; blind image deconvolution. I. INTRODUCTION Blur is a result of imperfect image formation process. It may be either due to the relative motion between the camera and the object or due to an out of focus optical system. Based on this blur is classified into motion blur and out of focus blur. Motion blur mainly occurs in a dim lighting environment where long exposure time is required. Motion blur can be modelled as the convolution of the sharp image u with the blur kernel k or point spread function (PSF). PSF refers to the extent to which an image of a point source is blurred by the motion blur. g = u ⊗ k + n (1) where g is the blurred image ⊗ is the convolution operator k is the blur kernel and n is the noise. The goal of deblurring is to recover sharp image from the input blurred image. Generally deblurring algorithms are classified into two blind image deconvolution and non blind image deconvolution. In blind image deconvolution blur kernel is known whereas in non blind image deconvolution blur kernel is unknown hence it is more difficult to solve. II. RELATED WORKS Recovering high quality image from a blurred image is a well studied problem. However constructing true image from a blurred image without artifacts is still a challenging problem. Fergus et al introduced a method for removing camera shake effects from photographs [1]. Their efforts were focused on kernel estimation and the estimated kernels seem to match the camera motion. But the recovered image often contains ringing artifacts. Shan et al. proposed a deblurring algorithm that suppress ringing artifact using Maximum a posterior approach (MAP) [2]. However, their method fails to deblur images with large size kernels since their algorithm incurs heavy computational cost. A fast motion deblurring approach using image derivatives was proposed by Cho and Lee [3]. They accelerated the sharp image estimation process by introducing a prediction step in the iterative deblurring process. Even though the algorithm ensures a fast processing its deblurring performance is slightly inferior to the previous methods. Ben Ezra and Nayar developed a hybrid camera design that uses a fundamental tradeoff between spatial and temporal resolution to obtain the camera motion information [4]. This information was used to obtain the Point Spread Function. Since it is a hardware approach it cannot be applied to general purpose video cameras.