An Improved Image Matching Model by Linear Transformation

This paper proposes an image matching model based on the linear transformation. The image geometrical transformations including translational, rotational and affine transformations are considered. Our model defines the linear transformation between two images, and takes the Sum of Squared Difference (SSD) of image intensities as the error function for image matching. By minimizing the SSD, the optimal linear transformation parameters are computed in an iterative manner. This paper also proposes three strategies to reduce the computation complexity and improve the convergence rate. The experiment shows that this algorithm can successfully match three images that relates to translation, rotation or affine transformations. The proposed stratiges significantly improved the convergent rate of image matching algorithm.