In this paper,translation, rotation and scale properties of 2-D affine transformation are discussed.Visual and IR image registration require a feature-based apporach. Establishing affinetransformation and extracting edge points as feature, this method calculates thecross-correlation between edges and searches out the relation of corresponding edgepoints. The maximum correlation principle is involved to adaptively determine the optimalregistration parameters k、θ、Δx and Δy. Experimental results using real visual/IR sensor image arepresented. It is shown that this model is effective in image registration.