International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169 Volume: 2 Issue: 5 1325 – 1331 _______________________________________________________________________________________________ 1325 IJRITCC | May 2014, Available @ http://www.ijritcc.org _______________________________________________________________________________________ Comparative Analysis of Patch Based Image Restoration Techniques for Diversified Field Images. Ms.Snehal Gandhi 1 , Prof. A. D. Bhoi 2 , Prof. A. L. Wanare 3 12 G. H. Raisoni Institute of Engineering and Technology, University of Pune, Pune, India 3 Dr. D. Y. Patil School of Engineering,University of Pune, Pune, India. snehalgandhi9@gmail.com, amol.bhoi@raisoni.net, anil.wanare.@dypic.in Abstract— Recently there has been considerable increase in the casual and commercial uses of image and video capturing devices. Apart from their applications in photography, the captured data are often inputs to sophisticated object detection and tracking, andaction recognition methods. Captured images are often not of desired quality and need to be enhanced by software. One of the major causes of the performance degradations for most methods is the presence of noise. In literature, many image restoration techniques exists for the reduction of noise from degraded image, but they usually do not succeed when applied to diversified fields degraded images with Speckle, Poisson, Gaussian and Salt & Pepper noise. So if an Image restoration technique works well for a particular type of image we cannot assure its performance for other type of image. Similarly if one technique works well in restoration of image corrupted with a particular noise we cannot assure its performance in presence of another noise. So in this paper, we provide performance analysis of state of art image restoration techniques i.e. patch based image restoration technique for various combinations of noise and diversified field images. Along with that a comparative result is drawn which gives the details of efficiency of all the image restoration techniques taken into consideration. In this paper we propose a new patch based image restoration scheme for the removal of noise. This new restoration technique is compared with the existing state- of-art patch based techniques such as K-SVD, FoE and Gaussian FoE. The proposed restoration technique is shown to outperform alternative state-of-the-art restoration methods with synthetic noise to diversified field images both in terms of speed and restoration accuracy. Keywords— Image restoration, optimization, synthetic noise, diversified field image, Patches of images. __________________________________________________*****_________________________________________________ I. INTRODUCTION Inrecent years, images and videos have become integralparts of our lives. Applications now range from the casual documentation of events and visual communication to themore serious surveillance and medical fields. This has led to anever-increasing demand for accurate and visually pleasing images.However, images captured by modern cameras are invariablycorrupted by noise.Image data obtained by camera sensors are generally contaminated by noise. Image data may be degraded by imperfect instrument, problem with the data acquisition process, and interfering natural phenomena.Similarly image is greatly affected by capturing instruments, data transmission media, quantization and discrete sources of radiation. Furthermore, noise can be introduced by transmission errors and compression. Medical images are used in many biomedical applications for diagnosis from x–ray, computerized tomography (CT)and magnetic resonance imaging (MRI).Similarly in geosciences scientists use remote sensing images to monitor planetary bodies, distant starts, and galaxies, so images used for these applications must be without the interference of the noise. Digital images are prone to a many of types of noise. Noise is nothing butthe errors in pixel values of the image, that do not reflect the true intensities of the real scene [1][2]. There are lots of types of noise which degrade the image. Each noise has its own source and its own characteristics. So if one image restoration technique works well for a specific type of noise it does not guarantee its performance in presence of other types of noise. So calibrating the performance of any image restoration technique with just one type of sample image and one type of degradation that to AWGN is not sufficient. So in our paper for comparing the performance of our proposed patch based image restoration technique with other state-of-art techniques we have chosen to take four noises into consideration. The noises taken for comparative analysis are Speckle, Gaussian, Salt& Pepper and Poisson noise. Similarly Evaluating the performance of the above four techniques on the basis of one sample image is also not correct so we have taken seven different images, divided in three categories i.e. Medical, Natural and Arial. Let us now discuss in brief the characteristics of noises that we have considered, Speckle is a characteristic phenomenon in laser synthetic aperture radar images, or ultrasound images. Its effects are caused by interference between coherent waves that, back scattered by natural surfaces, arrive out of phase at the source [3].Gaussian noise is an additive, which degrades image quality that originate from many microscopic diffused reflections leads to discriminate fine details of the image in diagnostic purposes [4].Impulse noise or Poisson noise in digital image is present due to bit error while source coding in transmission or introduced during the signal acquisition steps. Salt & Pepper noise can degrade the images where the affected pixel takes either maximum or minimum gray level [5][6].