International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169 Volume: 2 Issue: 8 2497 – 2502 _______________________________________________________________________________________________ 2497 IJRITCC | August 2014, Available @ http://www.ijritcc.org _______________________________________________________________________________________ Single image super resolution using compressive K-SVD and fusion of sparse approximation algorithms Aneesh G Nath Department of Computer Science and Engineering TKM College of Engineering Kollam, India aneeshgnath@yahoo.co.in Retheesh V V Department of Computer Science & Engineering NITK Surathkal, Manglore Karataka, India rethboby@gmail.com Abstract— Super Resolution based on Compressed Sensing (CS) considers low resolution (LR) image patch as the compressive measurement of its corresponding high resolution (HR) patch. In this paper we propose a single image super resolution scheme with compressive K-SVD algorithm(CKSVD) for dictionary learning incorporating fusion of sparse approximation algorithms to produce better results. The CKSVD algorithm is able to learn a dictionary on a set of training signals using only compressive sensing measurements of them. In the fusion based scheme used for sparse approximation, several CS reconstruction algorithms participate and they are executed in parallel, independently. The final estimate of the underlying sparse signal is derived by fusing the estimates obtained from the participating algorithms. The experimental results show that the proposed scheme demands fewer CS measurements for creating better quality super resolved images in terms of both PSNR and visual perception. Keywords- super resolution; compressive K-SVD;OMP;sparsity. __________________________________________________*****_________________________________________________ I. INTRODUCTION Super-resolution (SR) is the process of generating single high resolution (HR) image from one or more low resolution (LR) images. In traditional methods of SR the low resolution images that were captured with sub-pixel accuracy are used to solve the missing high-frequency information. But it is more challenging to recover this information from a single, low- resolution image. In many applications only a single low- resolution image is available and the Single image super- resolution (SISR) problem is particularly important in those situations. Compressed sensing (CS) is an emerging data acquisition technique which overcomes the limitations of Shannons sampling theorem. The motivating fact behind CS is that many natural signals are sparse or approximately sparse in a certain basis like wavelet, fourier etc. In many emerging applications, the abundance of data generated by the sensing systems due to high sampling rate demands data compression before storage or transmission. Compressed Sensing combines the sampling and the compression into a single process. CS data acquisition technique enables the reduction in the number of measurements required for recovery of sparse signals or compressible signals which is sparse on some suitable basis. Reconstruction of the signals from CS measurement is done using greedy or relaxation based algorithms. In a CS based image acquisition system it acquires less number of random linear measurements (pixels at a subset of sampling lattice) without first collecting all the pixel values. Proper preprocessing technique will enable the reconstruction of image from this incomplete data. The resulting image is an LR image which is suitable for reconstruction of its original HR image. In compressive sensing based single image super resolution, a low resolution input image plays the role of the compressive measurement of its corresponding high resolution image, and a proper dictionary which represents the high resolution image sufficiently sparse will make an accurate recovery of high resolution image. The problem of SISR is to obtain HR image X from its degraded Low-Resolution (LR) version Y, represented as Y = HBX + v (1) where H, B and v represent the downsampling operator, blurring operator and the additive noise respectively. In this recovery problem, we use CS based reconstruction method based on dictionary learning to generate the HR image. CKSVD an algorithm for learning a dictionary based on a given set of CS measurements, which is a generalization of the well-known K-SVD algorithm. More precisely, this algorithm is an iterative approach that alternates between sparse coding and dictionary update steps to minimize the error in representation of the CS measurements. The sparse coding stage is performed by fusing sparse approximation algorithms like OMP, SP etc. This paper is organized as follows. Section 2 describes the related works reported in the literature, Section 3 describes about the proposed super-resolution scheme including the dictionary training and reconstruction phases with the preprocessing step. Experimental results are shown in Section 4 and conclusions are drawn in Section 5. II. LITERATURE SURVEY SR reconstruction has been one of the most active research areas since the 1984 by seminal work of Tsai et al.[1] Many techniques have been proposed over the last three decades from frequency domain approach of Borman et al.[2] to spatial domain approach of Sung et al.[3], and from signal processing perspective to machine learning perspective. Conventional approach in super resolution is to generate a SR image from multiple low-resolution input images which are registered and