Fast Remote Sensing Image Registration Algorithm Using Spatial Masks

In this paper, a novel fast remote sensing image registration framework is proposed. The framework based on spatially masking reoriented key points patches generating a 64-dimension feature descriptor that results an enhancements for both memory allocation and processing time costs comparative to other state-of-art. The key points localized, corresponding patches retrieved at constant processing cost (patch size independent) using indexing techniques. Patches descriptors are constructed, filtered and matched for all key patches in a single execution step per each phase and a homography estimated from the filtered matched descriptors coordinates. The framework produces accurate and fast results with similarity transformation including translation, rotation and scaling, which is the dominant case of the remotely sensed image.