Comparing to the traditional multispectral images, hyperspectral images include richer and finer spectral information than the images we can obtain before. Theoretically, using hyper-spectral images should increase our abilities in classi-fying land use/cover types. However, when traditional classification technologies are applied to process hyperspectral images, people are usually disappointed at the consequences of low efficiency, needing a large amount of training data, and hard improvement of classifica-tion accuracy. In order to solve this problem, our atten-tion in the paper is focused on dimensionality reduction by feature selection or ex-traction. In this paper, we propose two new methods for feature extraction. The Two methods based on Frequency Space are Fourrier Spectrum Feature Extrac-tion and Wavelet Decomposition Feature Extrac-tion. We also compare with other feature extraction methods which have been developed from some other proposed papers. Finally, a practical AVIRIS data are analyzed to illustrate our discovery and test to show the efficiency of the new feature extraction methods.

This investigation uses feature matching which includes feature extraction, region descriptor and similarity assessment to initialize the correspondence for image registration. Shapes of the extracted regions appear to be in high frequency and low frequency in terms of shape irregularity. This irregularity, in general, is not predictable in the feature extraction stage Thus, we consider both possibility and then strive to investigate the suitable matching strategy. We propose three descriptors to characterize the region shapes, namely invariant moments, Fourier descriptors, and shape matrix. According to the matching matrix designed in this study, we screen out the erroneous correspondence. Then for these determined conjugate regions, centroids are used as registration control point in the initial stage. Following a coordinate transformation as prediction, fine matching is applied to generate final registration control point pairs.

With the progress of the informative technology, traditional methods of displaying the spatial information on 2D analogue maps have also evolved to 3D display in digital form. Virtual Reality, 3D City Models, etc. are, among others the examples of this evolution. As a result, how to automatically extract the man-made buildings from aerial images becomes a very important research topic in the field of the Digital Photogrammetry. The automatic building extraction always begins with the extraction of gray level edges. But the gray level edges detected by most edge detecting algorithms are often broken or irrelevant to building edges. In order to cope with these problems, other knowledge should be included to assist the search of the building edges. Digital Surface Model (DSM) is one of them. Digital Surface Model is the representation of the surface height information of the terrain, 'including the height information of buildings, trees and other fixed objects standing on the ground. Through proper analysis, approximate roof outline can be detected form the DSM. These outlines can then be used to assist the search of the building edges. In this research we assume that roofs are consisted of simple planar surfaces. Based on this assumption, surface normals which belong to the same planar surface should be have same orientations in the space. Therefore we can use the region growing technique to detect the approximate borders of that plane. The results of this research show that under proper conditions the approximate locations of building edges can be detected. These locations can then be used to screen out gray leve1 edges which do not belong to any buildings.

A test field consisting of 79 points has been established in the exterior of a building in National Chiao-Tung University, Hsin-Chu. The three dimensional coordinates of these 79 points are measured with a total station. The accuracy of the coordinates is within 1 cm. Al1 these points are chosen from the corner or the intersection of edges. No signalized points were used. A Wild P32 metric camera is used for a photogrammetric evaluation of this test field. A web page is established for documenting this test field. Both the point definition and the coordinates are provided through the net.