Research Of Related Correspondence Construction In Two-Dimensional Animation

Today,3D animation has become more and more popular. However, two-dimensional animation is still widely applied in relevant fields such as entertainment, advertising, education and so on due to unique art style and strong visual appeal. In particular, the realization of two-dimensional animation automation has more research value. Therefore, the study of two-dimensional animation especially correspondence construction has always been difficult and hot. Now, we study about the algorithm for correspondence construction of two-dimensional animation and its main contents are as follows:Our paper proposes two different matching algorithms, which are shape-based correspondence construction and skeleton-based correspondence construction. The former algorithm is based on the framework of constructing local patch using sparse coding and then whole alignment. Specifically, using the shape context to represent sample point of the outlier of the keyframe, then utilizing the LASSO model to combine with LARS algorithm to construct local patch and align these patches in a new feature space to achieve correspondence in clusters. The latter algorithm is using scale invariant transform feature to represent skeleton to construct correspondence. Firstly, it adopts Gaussian mixture model and iterative Graph Cuts namely Grab Cuts algorithm to separate foreground and background from scene. This is the prerequisite for our research in correspondence construction in two-dimensional animation. Then it obtains the object from the scene and use scale invariant feature transform (SIFT) to descried the skeleton of obtaining object. Finally, two objects can be detected by Graph cuts in the scene.Through experiments suggest our proposed algorithm is more effective. Moreover, the potential of our research for general application such as image matching, image retrieval, shape retrieval and two-dimensional animation and so on.