Super Resolution in Spatial and Wavelet domain

Abstract

Majority of digital imaging applications utilize high-resolution video or images for further processing. The requirement of high resolution video corresponds to two foremost application domains; enhancement of pictorial data for human translation and for programmed machine perception. Imaging acquisition devices generally restrict image resolution. Super-resolution is an idea according to which a high-resolution image or video can be produced by combining low-resolution series of images or videos of a sight. Of late, algorithms for single image super resolutions have been advanced, the most current research dependind single image or video Su R is based on texture hallucination, patch based up sampling and example based Su R, however problem of blur production and over smoothness persists. In this thesis an algorithm is proposed for SR of videos that utilizes the blend of interpolation alongwith wavelet transform technique. To enhance the efficiency of algoritham guided filters are added to preserve the edges and to retain the maximum information present in the video along the edges . It is notable that in process of generating high resolution output, the blurring effect arises in the section containing details that is mostly near bounderies. The proposed technique attempts to minimize the blur effect across boundaries. Proposed SR process consists of three main phases, Guided filtering for edge preservation, interpolation based magnification process and wavelet based edge boosting. The research concludes that the combination of these three techniques provides improved results both qualitatively and quantitatively.A comparison of this algorithm with other techniques proposed by other authors is also done to prove the effectiveness of the methods