| Summary: | 碩士 === 國立中央大學 === 土木工程研究所 === 92 === The orientation parameters are solved from the satellite header files information and adjustment model. Then adding digital terrain model is in order to develop relations between object space and image space. In this study, we use the groundel area images to match.
The research studies on image matching by variance and covariance components, and expects to improve the image matching precision. The study uses the best invariant quadratic unbiased estimator (BIQUE). In this way, it is the most important to segment the observations. The research uses the between-class variance to do it. It separates the observations into two parts, so we can get two variance components and one covariance components. The variance and covariance components have its accompanying matrices, so we have three accompanying matrices. To assign the variance and covariance components and to adjust relative weights through iteration until a steady parameter state is reached. After that, the least-squares image matching can use the weights in order to improve on a conventional stochastic model.
The study uses the Radarsat-1 image to test. After weighting in BIQUE, the RMS improves about 0.1 pixels in window sizes between 9×9 to 13×13 pixels.
|
|---|
