The Research And Implementation Of High-resolution Remote Sensing Image Matching

Recently, more and more high-resolution remot-sensing images emerge, which brings new opportunity and challenge to the field of remote sensing. On one hand, these high-resolution remot-sensing images make possible many new applications, such as Urban Planning, Graphics. On the other hand, due to the rich information that such images contain, more efficient method need to be devised to deal with these data.Image matching technology plays an important role in the interpretation of high-resolution remot-sensing images. For example, image mosaic, change detection, generation of DTM all need image maching as a prior . In other words, image matching can affect the final result of these applications to a large extent. Many image matching methods have been developed now, such as grey-based matching, feature-based matching and matching in frequency domain. These image matching methods have been widely applied in, for instance, fusion of medical image, 3-D reconstruction, object-tracking and so on. Image matching method is usually devised specifically in accord with its application's characteristic. In the field of high-resolution remot-sensing, an image matching method has to take into consideration the fact that high-resolution remot-sensing images usually have huge size and rich information of texture.Upon that, we present in this thesis an image maching method which can effectively control the propagation of error. This method first matches a few feature points which generate initial triangulation. After that, more feature points can be matched under the constraints of existing triangles. Traditional matching propagation methods with triangulation do not pay much attention to the maching error exsiting in the initial triangulation, so these initial errors will probably propagate in the whole matching process. In contrast, our method can prevent initial errors from affecting final result of image maching.