Image registration exploiting five-point coplanar perspective invariant and maximum-curvature point

Abstract

Image registrations have been a subject of extensive study over the last decade. They appear in numerous applications including Computer Vision, Pattern Recognition, Medical Image Analysis and Remotely Sensed Image Processing. We introduce a non-iterative geometric-based method to register an image using a novel set of geometric landmarks residing on an extracted 2D contours from the image. These landmarks are intrinsic and are computed from the differential geometry of the curve. We exploit the invariant properties of maximumcurvature points that are local and preserved under the affine and some perspective transformation. Geometric invariant exploits coplanar five-point invariant constructed from 4 consecutive landmarks and a centroid. This invariant is preserved not only in affine map but weak perspective map as well. To reduce the sensitivity of the landmarks to noise, we use a B-Spline surface representation that smoothes out the curve prior to the computation of the landmarks. The alignment is achieved by establishing correspondences between the landmarks after a conformal sorting based on derived absolute invariant. The experiments have shown that the purposed methods are robust and promising even in the presence of noise.