Abstract: A registration method for multi-spectral remote sensing image based on feature points was proposed. First,a two-degree regular mesh was formed on the image, and feature grids were chosen according to entropy and uniformity principle. Then feature points were detected by Forstner operate in feature grids. Due to the characteristic of multi spectral images, the course matching step was achieved by Cross-Correlation theory, and improved space distance constraint was used in fine matching step. Finally, registered image was obtained by affine transformation. RMSE was applied to evaluate the method. The experimental results clearly indicate that the approach we proposed is efficient with sub-pixel precision.

Key words: Remote sensing images,Fcature grid,Forstner,Course matching,Fine matching