Nikhil R. Kumbhar Int. Journal of Engineering Research and Applications www.ijera.com ISSN : 2248-9622, Vol. 5, Issue 5, ( Part -3) May 2015, pp.35-40 www.ijera.com 35 | Page Review of Use of Nonlocal Spectral – Spatial Structured Sparse Representation for Hyper spectral Imagery Restoration Nikhil R. Kumbhar*, Pratima P. Gumaste** *(Department of Electronics and Telecomm Engineering, JSPMs, Jayawantrao Sawant College of Engineering, Pune – 411028.) **(Department of Electronics and Telecomm Engineering, JSPMs, Jayawantrao Sawant College of Engineering, Pune – 411028.) ABSTRACT Noise reduction may be a vigorous analysis area in image method due to its importance in up the quality of image for object detection and classification. Throughout this paper, we've got a bent to develop a skinny illustration based noise reduction methodology for the hyperspectral imaging , that depends on the thought that the non-noise part in associate discovered signal is sparsely rotten over a redundant lexicon whereas the noise part does not have this property. The foremost contribution of the paper is at intervals the introduction of nonlocal similarity and spectral-spatial structure of hyperspectral imaging into skinny illustration. Non-locality suggests that the self-similarity of image, by that a full image is partitioned into some groups containing similar patches. The similar patches in each cluster unit sparsely delineate with a shared set of atoms throughout a lexicon making true signal and noise extra merely separated. Sparse illustration with spectral-spatial structure can exploit spectral and spatial joint correlations of hyperspectral imaging by victimization 3D blocks rather than 2-D patches for skinny secret writing, which collectively makes true signal and noise extra distinguished. Moreover, hyperspectral imaging has every signal-independent and signal-dependent noises, thus a mixed Poisson and man of science noise model is used. In order to create skinny illustration be insensitive to various noise distribution in numerous blocks, a variance-fitting transformation (VFT) is used to create their variance comparable, the advantages of the projected ways unit valid on every artificial and real hyperspectral remote sensing data sets. Keywords – VFT – Variance Fitting Transform I. INTRODUCTION The Hyper spectral images or imagery (HSI) has drawn several attentions from varied application fields. It provides information about each spectral and spatial distribution of distinct objects due to its varied and continuous spectral bands. During the imaging method, varied noises will be added into the Imagery, for time being, from imaging identifier and activity error. What is more, spectral, higher spatial, associated radiometric resolutions of hyper spectral imaging have junction rectifier to an increased impact of noises on the item detection and classification results. Several denoising approaches are applied for noise reduction of the Imagery, which include certain diffusion and filters (low pass, arithmetic filters), and wavelet shrinkage [1], [2], [3] [4]. Sparse illustration is associated in nursing rising and powerful tool for signal and image compression, noise removal, and classification. It gives information about signals supported the scantiness and redundancy of their representations. It usually assumes that signals like images, may be well resembled in terms of a linear combination of a couple of atoms in an exceedingly wordbook. Within the content of image restoration, the references in an exceeding wordbook talks to a collection of image bases. The low-dimensional nature of such illustration makes it applicable for process and analyzing these signals. As a result, it's incontestable compelling performance in several applications. Traditionally, noise reduction has been with success applied to one-dimensional acoustic signals and two- dimensional pictures. It faces some challenges once extended to HSI process and analysis. Normally, reduction of noise is done by using sparse representation on one – dimensional acoustic signals and 2D images. HSI being three dimensional images face problem to use the same method. Band by band processing techniques or pixel by pixel processing techniques are used in many noise reduction techniques. Hence due to such an approach there might be a correlation loss between bands or pixels. Now to overcome this loss some techniques working in sparse domain came into picture. In these techniques the important change is that the hyperspectral data is considered to be a three dimensional cube. In the approach [5] and [6], shrinkage of 3D wavelet or positive tucker decomposition is applied for denoising of HSI. The HSI data classification and compression can be done RESEARCH ARTICLE OPEN ACCESS