Registration of aerial images using moment invariants

Registration of aerial images is a necessary step as valuable information for studying; monitoring, forecasting and managing natural resources can be obtained from it. This paper presents an algorithm for registration of aerial images using orthogonal moment invariants. Initially reference aerial image and test aerial image are resized to same size and converted to gray scale and contrast enhancement is performed using nonlinear transformation. Further the corner points are detected in both reference and test images using Harris corner detector and are selected as control points for registration. A block of 21 × 21 pixels is selected around every detected control point in reference and test aerial image and invariant moments are computed for every block. Pearson correlation coefficient (PCC) and Euclidean distance (ED) are used for matching the moment invariants vectors representing corner points belonging to reference and test aerial image. The translation, rotation and scale present in the test aerial image with respect to reference aerial image are computed using similarity transformation. Performance of the proposed algorithm is tested on publicly available USC-SIPI Aerial Image Database. The average error for estimating rotation, translation and scale (θ = -5⁰, Δx = Δy = 50, S = 0.8) are θ = 1.27%, Δx =7.83%, Δy = 8.42%, S = 1.19% and average error for estimating rotation, translation and scale (θ = 10⁰, Δx = Δy = 25, S = 0.75) are θ = 2.31%, Δx =6.87%, Δy = 6.64%, S = 1.75% respectively.