Transformation-invariant image descriptors for change monitoring based on multi-modality imagery

An area-based multi-scale method for transformation-invariant descriptor extraction called multi-location feature saliency pattern (MFSP) is proposed in this paper in the context of image matching for change detection and monitoring. Multi-location image descriptors are extracted in salient circular fragments of variable size (scale), which indicate image locations with high intensity contrast, regional homogeneity and shape saliency. The MFSP is a set of relational descriptor vectors corresponding to a set of salient image fragments located in a neighborhood of a given feature point. The method proceeds without any image segmentation since the feature points are extracted by a fast recursive algorithm in a multi-scale manner analyzing circular high-contrast sub-regions of various sizes in every pixel location. The experimental results confirm the robustness of descriptor extraction by the proposed method and effectiveness of the multi-location feature saliency patterns for change detection and feature-based image matching.