Feature coverage for better homography estimation: An application to image stitching

The spatial distribution of feature points is known to affect the accuracy of homography estimation and hence the effectiveness of applications such as image stitching. This paper explores the use of spatial statistics as a metric for feature coverage, the distribution of features in an image. The coverages achieved by several feature detectors are compared and it is shown that detectors with higher coverages achieve better results in an image-stitching application. It is found that that SFOP is the most effective detector considered.