Robust Image Matching using Statistical Modeling and Geometric Similarity

We propose a robust image matching method using statistical modeling and clustering of geometric similarity between matching-pairs. Local feature matching is an uncertain process which may provide incorrect matches due to some causes that include among other factors, the uncertainly in feature location. Since the statistical modeling of the Log Distance Ratio (LDR) for outliers are significantly different from those of inliers. Although fast and efficiently, LDR has some weakness, especially related to the inability to take into consideration the uncertainly in the feature location and performance degrades when strong perspective transform. We add a method that clustering the similarity of geometric relationship. The proposed method robustly matches images, even with various kinds of transformation.