Image mosaic based on kernel regression

Abstract

Image mosaic technology mergers two over-lapping images into a seamless mosaic image. Classical parametric methods have been widely used to get mosaic pictures and show well effects. However, traditional parametric methods are highly rely on specific model and need a certain assumption, so in practice only can be used to a narrower aspect. In contrast with parametric methods, nonparametric method rely on the data itself, and due to its minimal assumption made on these data, nonparametric method is supposed to be applicable to wider problems than parametric methods. A new image mosaic method based on nonparametric method, say, kernel regression, is proposed, and experiments are presented which demonstrate the effectiveness of this method.