Scene matching by spatial relationships

Abstract

Scene matching is the process of recognizing two images as different views of the same scene captured using different sensor poses, and/or different types of sensors. In this work, each image contains the same objects and sensor pose parameters are not known. The spatial relationships among objects in the image, calculated using the histogram of forces (F-histogram) method, are used as matching elements. The degree of matching between two matching elements is calculated by comparing their F-histogram representations. Various geometric transformations are applied to the F-histograms during the comparison process to maximize the histogram similarity measure and to estimate the sensor pose parameters. The histogram similarity measure and the estimated sensor pose parameters are used as features in finding the best histogram correspondence map that matches the two images.