Research On Animation Copyright Recognition Technology Based On Character Action Features

With the development of Internet technology and the rise of major video platforms,the influence of animation works has been expanding.In recent years,China’s animation industry has developed rapidly and the animation industry chain has gradually formed.2021 saw a new record in the number of domestic animation releases,cumulative viewers and viewing hours.Animation is a comprehensive art expression which covers all aspects of literature,music,painting,film,etc.,and is very difficult to produce.A good animation work can usually generate great economic value.However,the repeated phenomenon of copyright infringement has become a stumbling block to the healthy development of China’s animation industry.At present,the copyright protection for digital animation works is not perfect,and China faces the dilemma of difficulty in proving the infringement in the judicial protection practice of animation works.Therefore,how to break the dilemma to improve the copyright protection of digital animation works has become an urgent problem to be solved.First of all,due to the simple texture composition and little detail information of animation works,the traditional feature point detection algorithm is not ideal.To address this problem,an effective feature point detection algorithm is given in this thesis.The method first segments the animated character in the key frame of the animation video;then performs edge detection to get the edge image of the animated character;and finally uses Shi-Tomasi’s corner point detection method to perform feature point detection on the edge image.The feature points extracted by this method are relatively stable and well distributed in the animated character,which is conducive to the extraction of subsequent animated character action features.Secondly,this thesis proposes an inter-frame action feature extraction method based on the pyramid LK optical flow method.Since animation works often use exaggerated expressions and animated characters have different shapes,it is often impossible to extract effective skeletal features.Therefore,in this thesis,some specific feature points on the animated characters are selected,and then the action features of the characters are calculated based on these feature points.First,the animation is segmented into shots according to the content of the video and key frames are extracted from each shot;then feature points are detected in the key frames and tracked using the pyramid LK optical flow method;feature points with tracking errors are removed by reverse verification,and then the inter-frame motion vectors of the remaining feature points are used as the local action features of the animated character.Finally,since most of the copies of animation will copy the content and action of the original animation and only change the debut character and drawing style,an animation copyright recognition method based on character action features is proposed for this kind of situation.Combining the feature point detection method and action feature extraction method proposed in this thesis,a set of synchronized original animation and copied animation videos are used as experimental data,and the local motion vectors in the two videos are extracted separately using the pyramid LK optical flow method,and through these local motion vectors,global motion features such as motion intensity,intensity mean,standard deviation and motion angle are calculated,and these features are mapped to vector space to get feature vectors of the two videos.The cosine similarity is used as the index to measure the similarity of the feature vectors,and the similarity is judged as plagiarism if it is higher than the set threshold.Although this method can effectively deal with the situation of animation plagiarism,it is sensitive to rotation and viewpoint change.Therefore,the copyright recognition method based on SIFT feature matching is proposed for the cases of pirated recording and pirated broadcasting.The SIFT feature matching algorithm has good robustness to scale,rotation,perspective,brightness,noise,blur and other changes caused by pirate recording and pirate broadcasting in animation works,and can accurately match the animation images.The experimental results show that the copyright recognition method based on character action features can effectively detect the existence of plagiarism in animation works.The method matches the action features of animation characters from frame to frame,so it has superior effect in identifying plagiarized animations compared with traditional image feature matching algorithms.In the copyright recognition experiments of three different categories of animation works,the detected plagiarism rates are improved by 24%,28.8% and 23.9%compared with the SIFT feature matching algorithm,and 26.8%,28.1% and24.7% compared with the ORB feature matching algorithm,respectively.