An optimization approach based on weight reassignment for remote sensing image registration and stitching

摘要

This paper presents an optimization approach for remote sensing image registration and stitching. A scale-invariant feature transform (SIFT) detector is used to extract feature points and obtain the homogeneous matrix between two images which have overlapped region. Remote sensing image has lots of details that may not evenly distribute. A method based on sub-block image entropy clustering to segment the image into large blocks is proposed. Each segment represents a unique scene type and will assign a weighted value to feature points. The weighted value reflects the amount of texture information and contributes to the improvement of optimization result. The experiment shows that the registration residual error can be reassigned and will be reduced in the region which has rich details.