Shape-aware Image Morphing

Image morphing technology is developed on the basis of computer graphics and mathematical image processing,which can realize the feature alignment of two input images through some methods,and generate the transition frames between two images,so as to realize morphing from the source image to the target image.This technology is widely used in the fields of science film making,animation design,3D reconstruction and so on.As the existing image morphing methods are difficult to deal with the problem of large geometric changes,this paper presents an image morphing technique which is based on shape interpolation and moving least squares deformation.In the algorithm,the features of the source and target images are at first manually specified using open or closed curves and then the source image is automatically transformed into the target image in a natural way.In order to retain original appearance and properties in the in-between objects,a barycenter-based shape model is designed to represent the visual appearance of a set of curves.Upon this model,a two-level shape interpolation is introduced to generate a sequence of intermediate feature curves which is appearance-preserving.During morphing,the intermediate feature curves are used to align the geometric features of source and target images via moving least squares deformation.Consequently,the original appearance and properties of the objects can be effectively preserved in intermediate images.We have applied our approach to a variety of examples to demonstrate its versatility and visual accuracy.In addition,this paper proposes a semi automatic image feature alignment method based on shape matching for the problem that inputting feature curves manually needs spend a lot of time and effort.In this method,users first specify a small number of feature correspondence points in the source and target images,and then the algorithm automatically completes the precise feature alignment between two input images.The core idea of this method is a progressive feature alignment mechanism.Upon this mechanism,the algorithm completes the precise alignment of corresponding features in source and target images from coarse to fine.Firstly,the algorithm takes user-specified feature points as constraint and uses the moving least squares deformation method based on similarity transformation to align the overall posture or shape of the feature objects in two images.Then,by means of the local neighborhood search method and shape matching technology,local detail alignment of corresponding features is realized.Finally,the global optimization method based on energy is used to complete the accurate feature alignment of source and target images at the pixel level.The experimental results show even when the geometric features of the source and target images vary greatly,using only a small amount of user specified characteristics corresponding points,this method still can achieve the precise feature alignment between two images,and therefore greatly reduce manual interaction workload.