Image matching method based on two-step switchable normalized deep neural network

Abstract

The invention relates to an image matching method based on a two-step switchable normalized deep neural network, and the method comprises the steps: analyzing and inputting a to-be-matched feature, and then adaptively outputting a matched matching pair through the training of a novel deep neural network. Specifically, a corresponding relationship between feature points in two views is given, and an image feature matching problem is expressed as a binary classification problem; and then an end-to-end neural network framework is constructed, and a two-step switchable normalization block is designed to improve the network performance in combination with the advantages of an adaptive normalizer for different sparse exchangeable normalization convolution layers and robust global context information of context normalization. The image matching method based on the deep neural network mainly comprises the steps of data set preparation, feature enhancement, feature learning and testing. The method can improve the matching precision.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to an image matching method based on a two-step switchable normalized deep neural network. Background technique [0002] Image matching is an important research field in computer vision. It is widely used in preprocessing in many fields, such as 3D reconstruction, simultaneous positioning and mapping, panorama stitching, stereo matching, etc. It mainly consists of two steps to build matching pairs and remove false matches. [0003] There are currently many image matching methods. They can be classified into parametric methods, non-parametric methods, and learning-based methods. Parametric methods are popular strategies for solving matching problems, such as RANSAC and its variants: PROSAC and USAC. Specifically, it first does random minimum subset sampling, generates a homography or fundamental matrix, then validates the matrix (whether it is the smallest possible subset...

AI Technical Summary

Problems solved by technology

However, these methods have two fundamental drawbacks: 1) they (parametric methods) cannot work effectively when the ratio of correct matches to total matches is low; 2) they cannot express complex models. Nonparametric methods mine local information for corresponding choice

However, they employ the same normalizer in all normalization layers throughout the network, which will lead to suboptimal performance

Images