Morphing using curves and shape interpolation techniques

Abstract

This paper presents solutions to the feature correspondence and feature interpolation problems in image morphing. The user specifies the correspondence between the source and the target images by drawing input curves on the features of the objects. The correspondence between these curves at the finest level (pixel level) is computed by optimizing a cost function. Based on this correspondence, the input curves are approximated by using Bezier curves. We represent the Bezier curves and the connections among them by using a "dependency graph". Feature interpolation is performed by interpolating the dependency graphs using the edge-angle blending technique. We also propose methods for controlling the transition rates of the shape and the color. We implemented the proposed algorithms in our morphing system which is based on the field morphing technique. From experimental results, our algorithms can generate a smooth morphing animation even when the objects in the source and the target images have different orientations. Also, the user's workload is reduced because the system includes an automatic feature correspondence computation.