基于对象识别与挖掘技术的视频水墨风格化算法研究

Nowadays, the video resource is greatly enriched, but the creation of Chinese ink painting animation is still lack of effective techniques and tools to take full advantage of ink style conversion of video resource. So based on existing research outcomes in terms of image processing, intelligent video analysis and recognition as well as simulative drawing of Chinese ink paintings, this project will focus on a series of algorithmic research in ink-painting stylization of video sequences. Firstly, intelligent video analysis techniques are applied for the extraction, tracking, mining and classification of video objects. Then, in combination to the extracted object-based motion parameters, ink-painting stylization of video with continuously rendering over frames is achieved. The main research topics include mining and classification of video object, automatic modelling and assessment of sketch and layout aesthetics based on image feature extraction and machine learning techniques, pen modelling for contour rendering of key frames, local stylization of key frames based on fluid diffusion and inter-frame continuous rendering based on motion features within the video. In addition, based on the outcomes of the project, an interactive system for intelligent ink-painting stylization of video will be developed. This will provide new concepts, technologies and tools for the production of traditional ink-painting animation, and results in broad application prospects in cultural and creative industries such as producing special effects in film and television programmes along with animation and gaming. Consequently, this project has significant and invaluable effects for the promotion of national culture and the protection of cultural heritage.

针对目前视频资源极大丰富，而中国水墨画动画的创作却缺乏有效的技术和工具来充分利用视频内容进行水墨风格化转换的问题，本项目提出基于对象挖掘和分类技术的视频序列水墨风格化算法研究。通过引入智能视频处理技术，首先对视频中的主要对象进行特征提取、匹配、数据挖掘和对象分类；在此基础上，结合所获得的视频体的运动特征参数，进行关键帧和帧间的视频水墨风格化渲染，实现水墨风格化结果在时空上的连续性和一致性。具体研究内容包括视频对象挖掘和分类算法、关键帧笔道形态及分布的美观度评估模型、关键帧视频对象的轮廓风格化算法、关键帧局部区域风格化算法、帧间连续风格化方法等。并在上述各项研究的基础上设计开发一个交互式的智能视频水墨风格化系统。该研究课题将为传统水墨动画的制作提供新的思路、技术和工具。其成果在影视特效制作，动漫游戏等文化创意产业具有广阔的应用前景。对于弘扬民族文化，保护文化遗产具有重要意义和价值。

根据该基金的研究内容和目标，我们重点对视频的特征提取、分类和水墨画图像的艺术风格分析和分类进行了研究，并尝试结合两者的内容进行风格化的研究。该项目基于视频、图像、人工智能等一系列交叉领域的知识展开视频水墨风格化等相关问题的算法研究和工程实践。首先对视频分析中最重要的分类识别问题展开研究，重点研究了其中的摄像机运动参数估计、特征提取、特征选择和分类方法，通过这些探索获得了视频对象识别问题中重要的特征和识别方法；然后结合先进的深度学习等机器学习方法，尝试以水墨图像数据库为数据集，针对它们的图像和笔道开展特征提取、图像稀疏特征提取、蒙特卡罗特凸集征选择方法、深度学习等方法的研究，实现水墨画的艺术风格分析、提取和学习；最后将视频分析、水墨画艺术风格学习和绘制的方法结合，探索视频水墨风格化的方法。研究成果方面，目前已经发表SCI期刊论文7篇，国际会议1篇，国内核心1篇。该课题的研究为水墨画艺术风格的分析和应用提供了新的思路和方法，未来在影视制作、美术动漫、文化宣传、包装、创意产业等方面有广阔的应用前景，对于弘扬民族文化，保护国家特色文化遗产有重要作用。