Emotional Semantic Recognition of Visual Scene in Flash Animation

Abstract

Based on the organization structure of the Flash animation files, we first use the edge density method to segment the Flash animation to obtain the visual scenes, then extract the visual features such as color and texture as the input parameters of BP neural network, and set up the sample database. Secondly, we choose a suitable model for emotion classification, use eight kinds of emotional adjectives to describe the emotion of Flash animation, such as warm, delightful, exaggerated, funny, desolate, dreary, complex, and illusory, and mark the emotion value of the visual scene in the sample database and so use it as the output parameter of the BP neural network. Finally, we use BP neural network with appropriate transfer function and learning function for training to obtain the rules for mapping from visual features of the visual scene to semantic space and, at last, complete the automatic classification work of emotional semantic of the visual scene. We used the algorithm to carry on the emotional semantics recognition to 5012 visual scenes, and the experiment effect is good. The results of our study can be used in the classification, retrieval, and other fields of Flash animation based on emotional semantics.