Information theoretic approach for template matching in registration of partially overlapped aerial imagery

Image registration is used in computer vision, medical imaging and remote sensing providing the ability to perform 3-D Reconstruction, Autonomous Navigation and Target Detection and Recognition Systems. Two of the more commonly used intensity based similarity measures in template matching for image registration are normalized cross correlation and mutual information. This works presents a novel information theoretic technique as a similarity measure for registration of partially overlapped aerial imagery. Furthermore, system level noise such as sensor noise, quantization noise, and impulse noise is modelled and injected into both the reference and unregistered images to evaluate the algorithmic performance in determining image orientation as a function of signal to noise ratio (SNR).