End-to-end image template matching method based on twin network

Abstract

The invention belongs to the technical field of image processing, and particularly relates to an end-to-end image template matching method based on a twin network. The method takes a template matching task as a classification regression problem to be processed and can better solve the problem of scale difference between a template and a reference image, and the robustness of template matching under a complex condition is effectively improved. The method has the following beneficial effects of 1, processing a template matching task as a classification regression task so as to better solve the problem of scale difference, and effectively improve the robustness of template matching under the complex condition; 2, combining a deep cross-correlation operation with a channel attention mechanism, providing a new cross-correlation operation for feature fusion so as to effectively improve the accuracy of template positioning; and 3, in the design of a loss function, using a DIOU for replacing a common IoU to construct the regression loss so as to enable the training process to be stable, accelerate the convergence, and meanwhile further improve the template matching performance.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an end-to-end image template matching method based on a Siamese Network. Background technique [0002] Template matching is to locate a given template image in a reference image through similarity measurement. It is a basic technology in image processing and computer vision, and is widely used in the fields of target recognition, medical image processing, and remote sensing. Due to different shooting times, angles, and imaging devices, there are often grayscale differences (even heterogeneous sources), scale differences, rotation differences, and viewing angle differences between the template image and the reference image. These differences bring great challenges to the template matching task. Similarity measurement methods used in traditional template matching methods include SAD (Sum of Absolute Differences), SSD (Sum of Squared Differences), NCC (Normalize...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems existing in the existing template matching method, the present invention proposes an end-to-end image template matching method based on Siamese network, which treats the template matching task as a classification and regression problem, and can better solve the problem of template and reference The problem of scale differences between images can effectively improve the robustness of template matching in complex situations

Images