A robust multi-stage information-theoretic approach for registration of partially overlapped hyperspectral aerial imagery and evaluation in the presence of system noise

Image registration is employed in a wide spectrum of fields ranging from medical imaging to remote sensing. This work presents a multi-stage hyper spectral image registration technique utilizing a hybrid feature-based and an area-based similarity measure for partially overlapped aerial imagery. The resulting selectively guided execution of similarity measures provides a reduction in search space for area based methods, reducing the computational cost of the proposed algorithm. The inherent statistical attributes of area-based methods are exploited through the sequential use of correlation and mutual information on physics-based features. The algorithm is evaluated in the presence of sensor uncertainty and system noise. This is critical as data acquired from different sensors may have varying level of sensor uncertainty that may reduce the performance of the post processing techniques. The registration parameters are assessed in the presence of sensor noise, quantization noise, and impulse noise. The three noise sources are modeled and injected into the images to determine the algorithmic performance as a function of signal to noise ratio (SNR).