Image Geo-Localization Based on Multiple Nearest Neighbor Feature Matching Using Generalized Graphs.

Area of Science:

• Computer Vision
• Geographic Information Systems (GIS)
• Machine Learning

Background:

• Image geo-location is crucial for various applications, but traditional methods struggle with feature ambiguity.
• Existing nearest neighbor matching can be unreliable, leading to incorrect correspondences.
• Global consistency in feature matching is essential for accurate geo-localization.

Purpose of the Study:

• To develop a novel framework for accurate image geo-location.
• To introduce a multiple nearest neighbor feature matching method using Generalized Minimum Clique Graphs (GMCP).
• To propose a robust distance function for enhanced global feature similarity assessment.

Main Methods:

• Extracting local features (e.g., SIFT) from query images.
• Employing GMCP for feature matching to ensure global consistency among selected nearest neighbors.
• Utilizing a Gaussian Radial Basis Function (G-RBF) based distance function for global feature similarity.

Main Results:

• The proposed GMCP-based method effectively selects globally consistent matches by considering multiple nearest neighbors.
• A novel G-RBF distance function enhances similarity assessment between global features.
• Experimental results on a 102k street view image dataset show a 10% improvement over state-of-the-art methods.

Conclusions:

• The novel framework significantly enhances image geo-location accuracy.
• GMCP provides a robust approach for achieving global consistency in feature matching.
• The proposed G-RBF distance function is effective for handling dissimilar global features in geo-location tasks.