Abstract
We present a novel, robust image feature detector. The proposed detector extracts highly stable and repeatable features based on the key idea of tracking multi-scale interest points for local structure-wise feature representation and evaluating the local corner signature (LCS) for estimation of a unique representative local structure with the strongest filter response in both spatial and scale domains. The experimental results and performance evaluation show that our feature detector has high repeatability and invariance to various geometric and photometric image transformations, and can be reliably used in image matching and recognition problems.
Animation Demo
Feature Detection Demo (Use ACDSEE to view it)
Feature Matching Demo (Use ACDSEE to view it)
Results
Detection Process:
Multi-scale Grouping:
Extracted Features:
Performance Evaluation:
