Automated wavelet based image fusion from video

In this thesis, we investigate a new method to produce high-resolution still images from a video sequence. The super-resolution problem, which refers to the creation of a restored high-resolution image from a set of low-resolution images using the redundant information between successive images, is considered here in a multiresolution framework. We use a wavelet transform approach to solve efficiently the different aspects (motion estimation, denoising, interpolation, deconvolution) related to the super-resolution problem. Our work focuses on improving the degree of automation and robustness of the super-resolution process by combining together different wavelet approaches to the above-mentioned aspects into a single framework. In the first chapter, we compare the classical algorithms for super-resolution, as well as some more recent multiresolution approaches. Chapter two provides an analysis of the wavelet theory and we identify some desirable properties required to solve the super-resolution problem in this context. The third chapter deals with the automated registration of colour images using a parametric global motion model. We also present an extension of the algorithm to detect and eliminate possible misregistrations. In the fourth chapter, the restoration aspects of the super-resolution problem are considered in the multiresolution framework. Chapter five gives a complete description of the algorithm. Finally, a study of the sensitivity of the method to several parameters is investigated in chapter six where the performance of the algorithm is evaluated using both synthetic data and real image sequences. Supplied by The British Library - 'The world's knowledge'EThOS - Electronic Theses Online ServiceGBUnited Kingdo