Document Image Registration Based on Geometric Invariant and Contour Matching

Registration method based on geometric invariant is efficient and accurate. However, it usually finds many legal but conflict transformations and determining registration transformation by geometric invariant error is not reliable. To enforce its robust in more general environments, this paper proposes a novel document image registration method based on geometric invariant and contour matching. Geometric invariant is used to find candidate transformations. The final registration transformation is determined through contour matching of two images. Experiments indicate its performance improvement.