The Method Of Auto-annotation For The Metairie Of Cartoon Character

With the fast development of cartoon industry, the demands of cartoon are increasing rapidly. However, there is a large distance between domestic research and international advanced research on computer assisted cartoon manufacture. Based upon the actuality of domestic cartoon industry, this paper gives a research on the automatic annotation of cartoon character material. Not only the framework of disposing character material is proposed, but also novel methods of automatic annotation for cartoon character material have been proposed. With the realization of the system on automatic annotation for cartoon character material, the validity of the methods has been proved. The original contributions of this thesis can be described as follows:According to the characteristic of cartoon character image, a description method has been presented in order to descript the local Invariant shape feature. Then based upon the feature, a matching strategy has been presented by using hausdorff distance. With the foundation of the work above, a novel framework for cartoon character image automatic annotation has been established. This framework can provide a retrieval result which has already been annotated. The annotation for input image could be obtained through the extraction of the retrieval result's annotations,According to the characteristic of cartoon character key-animetor a method for automatically annotating cartoon character video has been present by using improved HMM and star-skeleton model. According to this method, a behavior model of cartoon character is first been constructed by characters'pose, and then, using the mapping between human beings and cartoon characters, a novel characteristic of cartoon character video has been extracted by star-skeleton method. By using improved HMM to classify the video, the annotation of the input cartoon character video can be obtained.As a part of the computer assisted cartoon manufacture system, an auto-annotation system which can annotate large numbers of cartoon character material at one time is designed and implemented. The experiments have been presented to prove the validity of the system.