A Heterogeneous Image Matching Method Based on Aggregate Feature Difference Learning Network

Abstract

The invention discloses a heterogeneous image matching method based on an aggregation feature difference learning network, using a heterogeneous visible light-near infrared VIS-NIR data set, and using a Country subset as a training sample set, Field, Forest, Indoor, Mountain, Oldbuilding , Street, Urban and Water subsets as the test sample set; data preprocessing; design of aggregated feature difference learning network structure, including dual-branch feature extraction network, feature difference aggregation network, and two measurement networks; training aggregate feature difference learning network, based on The output of the two metric networks calculates the sum of two large-interval cosine loss functions respectively to obtain the final loss function of the network, and jointly optimizes the entire network; the test sample set is input into the aggregated feature difference learning network, and the matching label output by the metric network is used as the final predicted labels. The invention improves learning efficiency and matching precision, is more reliable, has stronger robustness, and has higher matching precision.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a heterogeneous image matching method based on an aggregation feature difference learning network. Background technique [0002] Predicting the matching relationship between image patches is very important in many computer vision tasks, such as image retrieval, person re-identification, image reconstruction, image registration, and object detection and tracking. Information complementation between heterogeneous images helps to further improve the accuracy of target detection, recognition and tracking. For example, in the case of good lighting conditions, visible light images can capture very rich detailed texture features. However, in poor lighting conditions, the quality of visible light images is very poor. Near-infrared images can make up for the shortcomings of visible light images that depend on lighting conditions, and clear images can also be obtaine...

AI Technical Summary

Problems solved by technology

The disadvantage is that it does not use the rich spatial structure information in the image block, which affects the effect of image matching and the accuracy of registration.

The disadvantage is that, first, only the features of the salient areas of the image are considered, and most of the image areas are ignored, which affects the accuracy of image registration

Third, using Zernike rotation invariant moments to describe features, the method of manually designing features has poor robustness, which makes the image matching and registration effect unstable and poor reliability

Images