Super Resolution Imaging – Wavelet based Interpolation using Hidden Markov Tree Model and Image Restoration by Cycle Spinning
Kurian Mathew, Sini Shibu · International Journal of Scientific Research · 2012
High resolution images are desirable in many applications such as medical diagnosis, remote sensing, moni- toring and surveillance. A brief survey of spatial domain super resolution enhancement methods that create high resolution beyond diffraction limit using several low resolution images is presented. Several low resolution images of the same scene at different sub pixel shifts are taken and registered on a common reference frame. A wavelet transform based interpolation is used. The image function is expanded in terms of a sealing function at a particular resolution jo and wavelet functions at different resolutions more than jo. The computation of wavelet coefficient is made easier by employing image pyramid. As wavelet coefficients show probability distribution relation of a Markov process and the statistical behavior of image signals can be suitably modeled using hidden Markov tree. Using Markov tree property we predict wavelet coefficient at finer scales or at higher resolution. This interpola- tion algorithm produces sharper images. For restoration of images we use cycle spinning and we get well defined edges and textures with high clarity.