An Image Matching Method Based on Multi-scale Neighborhood Deep Neural Network

Abstract

The invention relates to an image matching method based on a multi-scale neighbor deep neural network. This method analyzes the matching pairs that need to be matched as input, and adaptively outputs matching image pairs through a new type of neural network training. Specifically, given the matching data of two view feature points, an end-to-end neural network framework is designed. , the neural network formulates the image matching problem as a binary classification problem, which uses a specific compatibility-based distance to measure the distance between each matching pair, and then uses multi-scale neighbors to combine matching pairs and their neighbors into a graph. The method of the invention can fully mine the local information of the matching pair; compared with other matching algorithms, the method of the invention has achieved the most advanced performance in the benchmark data set and has better robustness.

Description

technical field [0001] The invention relates to computer vision technology, in particular to an image matching method based on a multi-scale neighbor deep neural network. Background technique [0002] Establishing reliable feature matching is a fundamental problem in computer vision, e.g., multi-label classification, panoramic stitching, and geometric model fitting. Finding robust feature matches mainly relies on two steps, namely, match generation and match selection. The first step is to initially generate a matching set using feature points. However, due to localization errors of local feature points and ambiguity of local descriptors, the initial matching is often unavoidably polluted by outliers. Therefore, the second step (i.e., selecting the correct match from preliminary matches) plays an important role in robust matching. [0003] Feature matching methods can be mainly divided into parametric methods, non-parametric methods and learning-based methods. Parametric...

AI Technical Summary

Problems solved by technology

However, these methods have two fundamental drawbacks: 1) they cannot work effectively when the ratio of correct matches to the total matches is low; 2) parametric methods are good at describing single geometric models, and it is difficult to express complex models (such as non-rigid match and multiple coincidence match)

Although NM-Net solves the local information mining problem of LGC-Net, NM-Net treats all neighbors with the same weight, ignoring the information between different neighbors

Images