A novel multi-frame super resolution algorithm for surveillance camera image reconstruction

This paper gives a loom towards the growing spatial resolution necessary to beat the limitations of the imaging technology in surveillance and security disciplines. It has been observed that metropolis cities worldwide invest huge sum of money in surveillance camera system but few are closely observing the benefits and the costs of those investments and to measure the overall impact of surveillance cameras on crime rates. The low resolution coupled with poor quality optics is not be enough to identify the subject of interest in crowd, from a distance, in bad weather and any other limiting factor. In this paper we have introduced multi-frame super-resolution technique that does not require explicit motion estimation and will be useful for producing imagery evidence that the police might reasonably accept as proof of someone's identity. Mostly the research is done in this area by taking a SR image and then after adding their own noise patterns where as our algorithm are working on actual LR images of surveillance camera and getting a SR image while removing the original blur and noise. Our algorithm requires the training set of Low resolution (LR) images from a still camera to produce High resolution (HR) image data and enhances it using anisotropic Diffusion and De-noising. In the image based representations, this technique of super resolution provides a great step towards resolution independence. The application of this method was successfully demonstrated for the restoration from a short low resolution set of images into a super resolved image. This super resolution algorithm works best when the Diffusion is applied and noise reduction filters are applied.