Fast image registration for multisensor fusion using graphics hardware

This paper presents a fast image registration using graphics hardware for image fusion of high resolution aerial images taken by different sensors. The GPU (Graphics Processing Unit) provides high-performance compared to its price and executes rapid computation by supporting parallel architectures. Fast image registration should be gone ahead to execute high-speed fusion of high resolution aerial images. Our approach is based on clustering techniques using parameter space, but we estimate the number of feature consensus pairs in the feature space according to change of parameter values instead of conversion over the parameter space. To extract feature points, we modified the filter which has been used for TNO fusion method based on image information and frequency. Furthermore, we reduce the search range of mapping parameters, and utilize multi-resolution methods to alleviate high computational cost.We implemented all processes for automatic image registration on programmable graphics hardware, the speed of GPU-based registration increased double compared to CPU's.