A Blur-Invariant Interest Point Detector Based on Moment Symmetry for Motion and Gaussian Blurred Image Matching

Abstract: Interest point detector is crucial to local feature-based image matching. However, lacking of robustness to strong blur is the fatal flaw of existing interest point detectors. As far as the authors know, all of the existing image matching methods fail to match a blurred image (caused by camera motion and out of focus, etc.) and a non-blurred image, even though blurred image matching is a critical task for many image/video applications. This article presents a blur-invariant interest point detector for blurred image matching. The proposed detector applies some blur-invariant image moments to detect a kind of special interest points from images. The special interest points are based on a new concept called Moment Symmetry (MS). These interest points are very robust to blur unlike traditional interest points based on corners or blobs. Experimental results show the proposed detector outperforms the state of the art interest point detectors for blurred image matching.