Registration Techniques, Remote Sensing Images Based On Wavelet Edge Extraction Of Feature Points

Accurate image registration is essential for the successful creation and interpretation of remote multi-image information products. However, the enormous increase in the volume of remotely sensed data that is being acquired by an ever growing number of Earth Observation satellites means that image registration now plays a more vital role than ever before.Firstly, existing image registration algorithms and their merits and drawbacks are reviewed, then, this paper presents a new feature-based combined with gray-based image registration approach using wavelet transform to extract edge feature points. The proposed approach is simple and effective, it can achieve an accuracy up to sub-pixels. It is a great improvement in efficiency as well as reliability for the image registration problem.This paper has such special points as:1. The proposed approach applies wavelet transform to extract a number of feature points as the basis for registration. And multi-scale edges of image are obtained. Each selected feature points is an edge point whose edge response is the maximum within a neighborhood. The results show that the edge has good property and can be used for matching of images. In order to suppress the effect of noise edges, a criterion called edge correlation is introduced.2. Cross-correlation combined with sum of absolute value of difference are used as correlation criteria to measure the similarity of any two feature-points. So, the reliability of matching result is improved.3. Here, generally matching is substituted by locally matching via manually locating two pairs of controls points in advance. The scheme resolves the problem of multiple peak value related with generally matching, meanwhile, speeds up the matching process.4. During the real matching process, we check each candidate pair in advance to see if it can possibly become a correct matching pair. Finally, based on the set of correctly matched feature point pairs, the accurate transformation between tow partially overlapping images can be decided.