Research Of Automatic Remote Sensing Image Registration Based On Feature

The main purpose of the image registration is to propose an image conversion method to eliminate or reduce the geometric distortion caused by the different imaging conditions such as different sensors, different perspective or different time, namely to achieve alignment on the spatial location. Image registration is widely used in the change detection, information fusion, the field of environmental monitoring, image stitching, object recognition, weather forecasts, map updates and the other applications. On the basis of a comprehensive summary and analysis of registration technology research, This paper focuses on the study of feature-based remote sensing image registration techniques, and proposes a variety of feature-based remote sensing image registration methods which are proved by simulation experiments. The main achievements and contributions are listed as following:(1) Accessing to domestic and international image registration related fields of large number of literature, the paper conducts a comprehensive summary and analysis of Image registration and its related technologies, and makes simulation experiments according to the existing registration methods.Paper also analyzes the applicability, advantages and weaknesses of the various methods. In this way,It provides a guideline for the subsequent researchment.(2) Aiming at resolution insignificant optical images, two feature-based automatic registration algorithms are proposed. One is the virtual feature connected with mutual-information-position-controlling-based registration algorithm and the other is SIFT and mutual information-based registration algorithm. Classic Harris feature point detection technology and the SIFT detection technology are introduced in detail. The virtual characteristics matching criteria and mutual information matching criteria are proposed. This two algorithms are simulated and compared from the angle of precision, running time, and the stability. (3) Aiming at the large resolution difference registration between panchromatic image with high spectral image, a typical region-based image registration technique is proposed. Different region-feature extraction methods are adopted according to different data characteristics of panchromatic image and hyperspectral image. To panchromatic image, the histogram proportional threshold segmentation method connected with mathematical morphology method is used to extract the appropriate typical region; to high spectral image, hyperspectral image segmentation based on the mixed pixel unmixing method is proposed. According to the characteristics of the typical regions, the perimeter ratio, the area ratio and the longest shortest center to boundary distance ratio are taken as constraint. Finally using affine invariant minimum distance principle, the paper establishes the similarity matrix to find the most matching regions.(4) On the basis of existing registration evaluation criteria study, a new joint histogram-based evaluation standard is proposed. The experiments prove that the joint histogram evaluation standard is easy to implement with intuitiveness and accuracy.