Generating 3D thumbnails for 3D contents

Abstract

Creating effective thumbnails is a challenging task because finding important features and preserving the quality of the content is difficult. Additionally, as the popularity of 3D technologies is increasing, the usage area of 3D contents and the 3D content databases are expanding. However, for these contents, 2D thumbnails are insufficient since the current methods for generating them do not maintain the important and recognizable features of the content and give the idea of the content. In this thesis, we introduce a new thumbnail format, 3D thumbnail that helps users to understand the content of the 3D graphical scene or 3D video + depth by preserving the recognizable features and qualities. For creating 3D thumbnails, we developed a framework that generates 3D thumbnails for different 3D contents in order to create meaningful and qualified thumbnails. Thus, the system selects the best viewpoint in order to capture the scene with a high amount of detail for 3D graphical scenes. On the other hand, this process is different for 3D videos. In this case, by a saliency-depth based approach, we find the important objects on the selected frame of the 3D video and preserve them. Finally, after some steps such as placement, 3D rendering etc., the resulting thumbnails give more glorified information about the content. Finally, several experiments are presented which show that our proposed 3D thumbnail format is statistically (p <0.05 ) better than 2D thumbnails.