Matching Images at Sub-pixel Levels Based on Directional Wavelet Transforms

The paper presents a new method for extracting feature points based on multi-scale direction wavelet and matching images at sub-pixel levels. Firstly the images were decomposed by directional wavelet transform and the feature points were extracted at multi-scales. Then the descriptor and feature vector related to each feature points were established using a isotropy modular. After that, matching feature points on different images used Euclid-distance in the two vectors. Then the matching results were refined using the geometric constrains for eliminating false matching points. Finally, the images readjusted at sub-pixel levels. The experiments show that the method is reliable and accurate for extracting feature points and matching images at sub-pixel levels.