Heterologous image registration method based on shape context and HOG features

Abstract

The invention relates to a heterologous image registration method based on shape context and HOG features. Aiming at the problem that information contained in a single-mode image is limited, the invention provides an infrared and visible light image registration method based on shape context and HOG features. On the basis of foreground detection of the Gaussian mixture model, contour feature matching is realized through a proposed method of combining shape context and HOG features, then matching is extended to the whole shape by using a TPS conversion model, a regularization and scaling feature iteration recombination corresponding relation and estimation conversion are used to reduce an estimation error, and finally an RANSAC algorithm is used to remove wrong matching points. Compared with an existing shape context method, the method has the advantages that the edge and contour feature information is combined, the registration error is reduced, and the registration rate and the registration robustness of the heterogeneous image are improved.

Description

technical field [0001] The invention relates to an image registration method, in particular to a heterogeneous image registration algorithm based on shape context and HOG features. Background technique [0002] Multi-sensor information registration fusion has important applications in military investigation, video surveillance, remote sensing information fusion, tumor detection and other fields. Information fusion technology can highlight target information and enhance the understanding of the scene, thereby improving the accuracy of the multi-sensor fusion system to perceive the target. But without accurate registration of images from heterogenous sensors, the quality of image fusion will be degraded. [0003] The process of image registration is to find the corresponding relationship between the same position points in the two images according to the image space and gray scale changes, and provide high-definition, high-quality, and accurate target information for image fu...

AI Technical Summary

Problems solved by technology

Based on the geometric error calculation, the corresponding geometric transformation (translation, rotation, scaling, etc.) is performed on the image. By comparing the calculated images of various methods to be evaluated with the geometrically transformed images, the closer method has better accuracy. ; Based on image features, Bilodeau et al. proposed a local evaluation method based on the target coincidence rate, but in practical applications, it may not be possible to effectively evaluate the registration method because there is no selected target in the image; image-based Grayscale, the normalized mutual information index (NMI) between images is used as a global index, but the problem of local extremum still appears in multi-sensor image registration; based on subjective judgment, different observers use multiple registration This method is intuitive and convenient, but there is a certain instability in the evaluation results of different groups of people.

Images