A Multi-temporal SAR Image Registration Method Combining Stable Points and Feature Points

Abstract

The invention provides a multi-time-phase SAR image registration method combining a stable point and a characteristic point. The method comprises the following steps of carrying out correlation consistency (CS) area detection, and according to SAR image complex data, calculating Coherent Scatters of an image and carrying out SAR-FAST angle point extraction; calculating an intersection of a CS area position and a characteristic point position to acquire a stable characteristic point, and using a SIFT descriptor to describe; in a characteristic coarse matching phase, using a KD index tree to match a descriptor and then rejecting a mismatching point based on a RANSAC algorithm; and during accurate registration, using Powell to search optimal mutual information position so as to carry out accurate registration. In the invention, through using a mode of combining the stable point and the characteristic point, SAR image registration is performed so that high registration precision is acquired.

Description

Technical field [0001] The invention belongs to the field of image processing, and relates to SAR image registration using feature extraction and matching, in particular to a multi-temporal SAR image registration method combining stable points and feature points. Background technique [0002] SAR image registration is to align multiple SAR images with different imaging times, different angles, and different sensors. It is the basis for image interpretation such as change detection and image fusion. In optical and SAR image registration, the registration method based on feature points is mostly used. Feature-based methods are not only less computationally expensive, but also unlike line features and area features that require high scene content, and are not like gray-scale-based methods at the cost of computational complexity. However, due to the large amount of speckle noise in SAR images, the point features applied in SAR image registration generally pay more attention to overc...

AI Technical Summary

Problems solved by technology

At present, in the methods of multi-temporal SAR image registration, they are all sparse matching algorithms based on feature points. When extracting feature points, only the suppression of multi-spot noise is considered. For the unstable points that appear in multi-temporal SAR images The removal of , so far no one has achieved

Images