Oblique remote sensing image matching based on improved AKAZE algorithm
Yuxuan Liu, Chaozhen Lan, Fushan Yao, Lin Li, Canhai Li · 2016
In order to address the matching problems of oblique remote sensing image with viewpoint change, geometric deformation and radiometric distortion, this paper presents a new point-based matching method which conducts feature extraction by building nonlinear space on simulation images derived from a full-range resample of the origin image. All distortions caused by the position change of camera are first modeled by different tilts images; then the feature points are localized by improved Accelerated-KAZE (AKAZE) algorithm, namely the feature points are detected in nonlinear space constructed by Fast Explicit Diffusion (FED) and variable conductance function, and the extracted feature points are described by improved SIFT descriptor; finally, Euclidean distance is used as similar metric to determine the correspondences and Random Sample Consensus (RANSAC) algorithm is employed to eliminate the false matches. In experiments of oblique images and UAV images, the number of feature points extracted by the proposed method is biggest, the effective correspondences are most and the correct matching ratio is highest among SIFT, ASIFT and our proposed algorithm. Thus, our proposed algorithm can be successfully used in oblique remote sensing image matching.