Using Geometric Blur for Point Correspondence

In computer vision applications, point correspondence plays an important role. Here, we present a technique called geometric blur to find point correspondences between two different images, even in the presence of affine distortions. We compare the results of this technique with other prevalent techniques for finding point correspondences, such as SSD (Sum of Squared Differences) with uniform Gaussian blur. Experimental results are shown on various face images and other object images. Index Terms — point correspondences, geometric blur, interest points, object detection and recognition, SIFT