Feature extraction and image registration of color images using Fourier bases

This paper presents an approach to feature point matching between widely separated views of color images that have undergone affine transformations, color contrast and illumination changes. Firstly, we present a KLT like feature extractor for color images that uses Fourier amplitude measure to estimate the derivatives. Next we define Fourier invariants for matching feature points under large affine image deformations and illumination changes. Finally, the process of getting correct matches is enhanced by using the kernel density estimation of the color information in the neighborhood of feature points. Test image mosaics are generated to evaluate the proposed technique. The results justify the approach adopted.