A Novel Spatial Blind Super Resolution For Video-Sequences

Abstract

In Digital image processing the super-resolution of images is a growing technique because of its simple structure with high accuracy as well as reliability is more. As motion fields are having a very high complex nature of motion, so there is need to use the application of super resolution for real life video sequences. The proposed work is implementation for spatial resolution improvement by a novel technique known as blind super resolution(SR),while we unknown about the parameters like  noise statistics, point spread function as well as motion fields. In this we estimated the blur by multiscale process but before that we have to upsample the frames with the help of nonuniform interpolation super-resolution. In image if we study details we can know that most of the information present at the edges.So, there is need to first get proper edges of an image. The blur estimation methodology is applied on the few edges but as iteration goes on it is applied to almost all edges of image obtained by gradient. We got more faster convergence in this technique by adopting a pixel domain analysis rather than the filter domain analysis. No of pixel based techniques are there but we preferred Huber-Markov random field because of its very high resolution outputs with preservation of the edges and fine details. It is having two important terms like fidelity and regularization which are analyzed before applying the random field. The term ‘fidelity’ is continuously applied weighting by application of masking and the main aim of applying masking is to avoid the inaccurate motion based artifacts. Advantage of proposed work are it can handle complex motion problems, any deformable regions will be estimated accurately, efficient under different brightness condition, detailed structure is obtained as well as it can be applied to fast moving objects. The results obtained are analyzed by subjective and objective analysis to show its state of art over the existing techniques.