Multi-resolution image fusion using multi-scale estimation framework

Recently, we have been experiencing remarkable advance in remote sensing technology and it allows us to capture large classes of natural processes and phenomenon at different resolutions with confident levels of qualities. For example, data acquisition over specific topographic environment by employing high spatial resolution sensor is extremely useful in monitoring detail physical and biological processes of the Earth's surface. As data acquisition techniques become sophisticated using more accurate sensing devices, the demands for processing obtained data sets are more diverse and complex.;In this dissertation, we develop data fusion methods to process image sets obtained by heterogeneous sources at different resolutions. This fusion scenario is based upon the idea that tries to combine image sets differing resolutions by employing robust and efficient signal processing scheme like multi-scale estimation framework. Since the data collection processes have been performed by different methods and for different purposes, merging process is not a trivial task. Despite of this technical difficulty, real world remote sensing applications require information that is insufficient to be interpreted by a single sensor measurement. Since individual sensor employed is operated in certain acquisition geometries (e.g. altitude, distance or viewing angle), this turns disparate coverage and accuracy characteristics on obtained data. A number of attempts have been made to combine data from different sensors, but existing methods are often empirical based [Sorenson, 1970]. Successful data fusion becomes especially difficult when the sensors involved have significantly different acquisition methods, wavelengths, and resolutions.;To overcome this difficulty, this study shows the efforts to build robust image processing framework for combining (fusing) complementary multi-resolution image data sets. The contributions of this work are summarized as: (1) constructing statistically optimal fusion method utilizing multi-scale estimation framework such that one can efficiently obtain fused image estimate and its confident measure (uncertainty) at desired resolution, (2) developing new multi-scale fusion techniques to find solutions for computationally challenging fusion situations, and (3) extending multi-scale estimation techniques to generate both visually and analytically improved fused image at the finest resolution by mitigating pixel blocky artifacts commonly arising when the resolutions of the images being fused are differed by large orders of magnitudes and image pixel voids at fine-scale are severe.