Optimization of registration accuracy of distorted super-wide field-of-view infrared images based on global selection preserving matching with chunking

Super-wide field-of-view infrared images have attracted much attention in the military and civil fields due to their wide detection range and anti-interference ability, but their significant nonlinear distortion seriously restricts the accuracy of traditional alignment algorithms. The existing methods are mostly for ordinary field-of-view images which ignore the interference of super-wide field-of-view distortion on feature matching, and the sparseness of infrared image texture further increases the registration error. In this paper, we propose a global selection preserving matching method with chunking, which extracts the initial matching points by chunking the image to reduce the local distortion, combines with speeded up robust features, and rejects the false matches using the dual mechanism of global geometric consistency filtering and local triangulation verification. Experiments show that compared with five mainstream algorithms, our method on six pairs of super-wide field-of-view infrared images significantly reduces the registration error and improves the target splicing match, which verifies the key role of uniformly distributed correct matching points for the registration of distorted images. This research provides a new idea for high-precision super-wide field-of-view infrared image registration.