A hough voting strategy for registering historical aerial images to present-day satellite imagery

Abstract

In this paper we present an approach for the georeferencing of historical World War II images by registering the images to present-day satellite imagery, with the aim of supporting the risk assessment of unexploded ordnances. We propose to exploit the local geometry of corresponding interest points in a Hough voting scheme to identify the most likely transformation parameters between the images. Our method combines the evidences from local as well as global correspondences and uses a spatial zoning rule to establish solutions with preferably uniformly distributed correspondences. An experimental evaluation is conducted on a set of 42 pairs of historical and present-day images and reveals the outstanding performance of our method compared to state-of-the-art image matching and registration algorithms, including commonly used hypothesize-and-verify and graph matching methods.