Removing parallax-induced false changes in change detection

Accurate change detection (CD) results in urban environments is of interest to a diverse set of applications including military surveillance, environmental monitoring, and urban development. Although CD approaches with hyperspectral imagery exist, there is no comprehensive framework that describes existing approaches and provides a foundation to generate new ones. This work presents a hyperspectral CD (HSCD) framework. The framework uncovers the need for HSCD methods that resolve false change caused by image parallax. Image parallax is the apparent motion of stationary objects due to differing viewing geometries.;A Generalized Likelihood Ratio Test (GLRT) statistic for HSCD is developed that accommodates unknown mis-registration between imagery described by a prior probability density function for spatial mis-registration. Using a normal prior distribution for the mis-registration leads to a fourth order polynomial test statistic that can be numerically minimized over the unknown mis-registration parameters. A more computationally efficient closed-form solution is developed based on a quadratic approximation derived through a second order Taylor series expansion and provides comparable results to the numerical minimization for the investigated test cases while running 30 times faster. The results indicate an order of magnitude reduction in false alarms at the same detection rate for synthetically mis-registered test data, especially in image regions containing edges and fine spatial features. Investigation of the model parameters is also assessed, and the potential of the derived method to incorporate more complex signal proccessing functions is demonstrated by the incorporation of a parallax error mitigation component. The GLRT statistic is extended to account for parallax errors and the visual results demonstrate the reduction of false alarms; also evident in test statistic histograms and signal-to-clutter ratio (SCR) computations.