基于海量图像的真实感故事可视化生成  

Realistic Story Visualization from Online Photo Collections

作　　者：孙迪 詹睿[1] 张加万 潘刚[1] Sun Di;Zhan Rui;Zhang Jiawan;Pan Gang(School of Computer Software,Tianjin University,Tianjin 300350;School of Computer Science and Information Engineering,Tianjin University of Science and Technology,Tianjin 300222)

机构地区：[1]天津大学软件学院,天津300350 [2]天津科技大学计算机科学与信息工程学院,天津300222

出　　处：《计算机辅助设计与图形学学报》2018年第8期1479-1489,共11页Journal of Computer-Aided Design & Computer Graphics

基　　金：天津市哲学社会科学规划项目(TJSR15-008);国家社会科学基金(15XMZ057)

摘　　要：通过对故事可视化领域进行深入研究,提出一个完整的故事可视化系统.首先利用自然语言处理对故事脚本进行解析,自动形成场景单元及关键字,根据关键字从互联网上搜索相应的图像序列;然后提出一种改进的前景自动提取算法,并分别采用不同的过滤策略获取所需的候选前景和背景图像;最后根据用户选定的图像合成单幅场景,并采用一致性优化算法来保持场景之间的连续性与一致性.实验结果证明,该系统能够智能地辅助用户生成故事图像序列,增加其认知和创作乐趣,有效地减轻用户创作的负担和约束,而对用户本身无需任何艺术背景要求.文中工作可用于儿童认知以及导演创作等应用领域.A complete framework used for creating illustration of a given story is presented in this paper.Guided by a short story and simple sketches, users can create visual story representations from online photocollections. Based on natural language processing method, the framework decomposes a given input storyinto suitable units, and constructs meaningful keywords by which the corresponding images can be automaticallydownloaded from the Internet. An automatic image extraction method based on saliency region isposed for image foreground filtering. Besides, a consistency optimizing method between neighboring scenesis adopted to adjust inconsistent problems. Experimental results indicate that our framework can be used byany user to produce personalized picture series with specified story, even without art background. The workcan be used for children＇s cognition, director＇s creation and some other fields.

关 键 词：讲故事 故事可视化 海量图像 图像过滤 显著性提取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]