Quantitative analysis of resolution synthesis
Ramez Yoakeim, David Taubman · 2005
We address a number of gaps left by recent work on resolution synthesis image interpolation, including the validily of some assumptions which we examine and verify. Thc relative merit of a mixtore approach versus a maximum likclihood approach, the impact of thc classification proccss, and the impact afthe numbcr of classes on thc performance of the interpolator are all investigatcd. We also cnamine the suitability of the underlying statistical models. We propose a modificd synthesis component, based on the discrete wavelct transform, as an altemativc to the non-overlapped block synthesis proccss described in previous work. We also introducc a number of measures to significantly improve the computational cfficiency and suitability of RS for automated, unassistcd classification training.