Normalized cross correlation (NCC) registration method of self-adaptation threshold

Abstract

The invention provides a normalized cross correlation (NCC) registration method of a self-adaptation threshold. The NCC registration method comprises the following steps: an initial NCC threshold value is selected for registering feature points in two pictures; the registered feature points are utilized for obtaining an affine transformation matrix between the two pictures; affine transformation is carried on to-be-registered pictures by utilizing the obtained affine transformation matrix; mutual information entropy between the to-be-registered pictures after affine transformation and standard pictures is obtained; by utilizing the Powell optimizing iterative algorithm with a trial-and-error method and a golden section method, mutual information is used as search criterion to carry out self-adaptation iteration of the NCC threshold; the Powell algorithm is utilized for changing an NCC threshold value constantly and carrying out iteration; after the iteration is finished, the largest mutual information entropy is found and a corresponding NCC threshold value at the moment is obtained; and the obtained NCC threshold value is used as an ultimate threshold value for carrying out registration of the feature points. The NCC registration method is based on image registration of the feature points and used by matching with the feature point detection algorithm, and therefore the purpose of image registration can be achieved.

Description

technical field [0001] The invention relates to a digital image processing method, in particular to a registration method in the field of image algorithms. Background technique [0002] The image registration algorithm based on feature points is a common image registration algorithm. The advantage of this algorithm is that it can use the registration feature point pairs in the two images to be registered to directly obtain the transformation relationship between the two images. Implement image registration. The existing commonly used feature point detection algorithms generally include corner point detection algorithms such as: susan, mic and harris corner point detection algorithms, etc., and use the SIFT algorithm in the invariant technology. Different feature point detection algorithms have their own advantages and disadvantages. It is very necessary to choose a Harris corner detection algorithm with rotation invariance and good detection performance as the feature point...

AI Technical Summary

Problems solved by technology

Compared with the SIFT feature description word registration algorithm, the NCC registration algorithm has a much smaller calculation amount, and the registration effect is relatively ideal with low complexity.

However, due to the fixed threshold of the original NCC registration algorithm, it will lead to misalignment of feature point pairs or too few registered feature point pairs.

Images