An efficient technique for subpixel accuracy using integrated feature based image registration

Abstract

Image registration which is frequently required in Medical, Computer vision and remote sensing field is used to align two images geometrically. This paper presents efficient method for providing speedup and more accuracy in compared to current state of the art existing methods. This paper focuses on Feature detection using Harris detector which gives best result based on performance and has firm invariance to scale, rotation angle, illuminance and noise compare to other feature detection algorithms. Also Feature vector is generated for each detected interest point using Histogram of gradient. Generated feature vector are matched in both image using Best bin first (BBF) to generate accurate Interest point pair and to align the sense image. An additive step is applied to remove outliers and to improve registration accuracy. An experimental result shows that the present method using Harris detector, Hog descriptor, BBF and RANSAC provides higher precision and also it is more robust.