Image Super Resolution and Reconstruction Using Sparse Representation

Abstract

Quality of an image plays a main role in cameras, image enhancement, image reconstruction, and in the many latest technologies. The quality of the image is increased by the process of image super resolution where we access the different techniques.

One of the best technique for a single image super resolution is depends on the sparse signal illustrations, which gives the better results and robust to noise when compared to the previous methods. Most fundamental method as super resolution image reconstruction where super resolution image requires lot of lower resolution images. While the simple interpolation techniques represented as a Bilinear also called as Bicubic interpolation that generates overall soft images along with jagged and ringing artifacts. Machine learning techniques were also accessed for image super resolution but these techniques require a large database which in terms of millions which are very expensive. To overcome all this problems, we go for a new technique which truly rely on the sparse illustration. “Sparse”- a statement of the input signal as a linear mixture of base components in which huge numbers of the co efficient are zero. This approach tells us that image blotches can be written as sparse linear mixture of elements taken suitably from over-complete dictionary. Here we take the sparse illustration for every blotch in the lower –resolution image, and the coefficients obtained in this are accessed to retrieve higher-resolution image. Here we access two dictionaries lower and higher resolution dictionaries.

The above method is very sensible to the considered model of data and noise those automatically bars their access. To overcome these short comings new technique knownasL1norm minimization with the robust coordination which tackle various data and noise models lies on a primary bilateral concept which are less expensive in computational part and strong to noise that gives sharp borders in images and this method is superior than the remaining methods. The above traditional super resolution methods accessL1and alsoL2statistical norm estimation which are very sensible and bars their utility, to avoid this we go for the new technique deals with the stochastic coordination method of Bayesian MAP estimation which minimizes the cost function. Huber norm is managed to remove artifacts from the image and to boost the rate of convergence.