GPU and cascade Hash-based quick image SIFT feature matching method

Abstract

The invention discloses a GPU and cascade Hash-based quick image SIFT feature matching method and belongs to the field of computer vision. The method comprises the steps of establishing a three-stage exchange mechanism of a GPU, a memory and a hard disk according to size limitation of a global display memory of the GPU and size limitation of the memory; by adopting an improved GPU parallel reduction method, performing Hash mapping of two different encoding lengths on all SIFT feature points of an image; proposing and using an upper triangular matrix block reading method; performing coarse screening through a local sensitive Hash algorithm; performing fine screening on candidate points by calculating a Hamming distance; and finally finding a most suitable matching point by calculating Euclidean distances between the screened points and to-be-matched points. By utilizing an asynchronous parallel method of a CPU and the GPU, matching result data of an image pair is copied to the memory from the display memory of the GPU and stored in the disk during GPU operation; and the matching time of the SIFT feature points is greatly shortened and the feature matching also can be finished in a short time under the condition of massive data.

Description

technical field [0001] The invention belongs to the field of computer vision, and more specifically relates to a fast image SIFT feature matching method based on GPU and cascaded hashing. Background technique [0002] Image feature matching is the corresponding relationship of feature points between images, which has a wide range of applications in image recognition, image registration and 3D reconstruction. The SIFT feature has scale invariance and rotation invariance, so it has strong anti-interference ability and high matching accuracy, so it is a commonly used feature for image feature matching. However, each SIFT feature point is represented by a 128-dimensional vector. The similarity measure of the SIFT feature point is the Euclidean distance. In the case of a high-dimensional vector, the Euclidean distance calculation time is longer, and the linear search method matches between two images. The computational complexity is O(N 2 ), where N is the average number of fea...

AI Technical Summary

Problems solved by technology

However, in the case of a large number of images to be matched, such as 3D reconstruction of large scenes, some scenes require pairwise matching between images, then there are order of magnitude of matching pairs to process, where N p is the number of images, that is, in the case of massive image matching tasks, the cascade hash matching method still requires a lot of computing time

Images