Gaussian Affine Feature Detector

TL;DR

This paper introduces a Gaussian-based method for extracting affine shape features from images, offering a non-iterative, accurate, and robust approach suitable for various challenging conditions.

Contribution

It presents a theoretically optimal, non-iterative solution for affine feature detection using Gaussian signals, improving accuracy and robustness over conventional methods.

Findings

Achieves or outperforms existing methods in accuracy, speed, and stability.

Effectively detects small, long, or thin objects.

Performs well under low contrast, blurred, or noisy conditions.

Abstract

A new method is proposed to get image features' geometric information. Using Gaussian as an input signal, a theoretical optimal solution to calculate feature's affine shape is proposed. Based on analytic result of a feature model, the method is different from conventional iterative approaches. From the model, feature's parameters such as position, orientation, background luminance, contrast, area and aspect ratio can be extracted. Tested with synthesized and benchmark data, the method achieves or outperforms existing approaches in term of accuracy, speed and stability. The method can detect small, long or thin objects precisely, and works well under general conditions, such as for low contrast, blurred or noisy images.