Remotely sensed Images Registration Based on Wavelet Transform Using Affine Invariant Moment

This paper presents an accurate and automatic feature-based method for registration of remotely sensed images. The proposed method uses the redundant wavelet transform to obtain feature points. Firstly, modulus maximum of wavelet coefficients of reference and sensed images are detected based on multi-scale multiplication, respectively. Then, affine invariant moments are calculated using a circular neighborhood area centered on the detected modulus maximum Point. After that, corresponding feature points are detected by minimum distance rule with the threshold in the Euclidean space of the invariants. Finally, the transformation parameters are obtained by least square method. Also, the center spline function is used as interpolation function when wrapping the sensed image to reference image. Experiment applying our proposed method on Landsat TM remotely sensed images prove our proposed method effective.