Intelligent Processing of Visual Media

该项目涉及数字几何、图像和视频序列处理等多个领域，总投资2400万元（约1，600，000）由中国科技部。调查的主题将包括形状，图像和视频理解，使用神经网络和其他基于人类视觉系统结构的思想，重点是高性能系统开发处理几何和图像和视频理解所需的数学工具，允许计算机学习它们所看到的事物，然后对稍后展示给它们的新例子进行分类。这里的关键问题是存在于图像中的大量像素中的大量信息，在视频中更是如此。如何使用人类视觉系统的类比来帮助开发能够理解图像和视频的系统。这包括如何从小细节中分离出大画面，在不同的几何尺度上使用理解，以及如何制作可以自动聚类相关信息的方法。进一步的关键问题是如何获取和表示几何和视觉信息，以便有效和高效地处理这些信息，以回答有关内容的查询。几何形状，图像和视频的关键特征如何表示，存储，压缩和加密。特别重要的是数字版权管理问题，包括在视觉数据中嵌入不可见的水印以证明所有权，以及为几何模型、图像和视频提供内容感知安全的能力。其他目标包括捕获和编辑几何模型、图像和视频的新方法，包括视频混合。

The proposed research covers a very wide range of topics in processing of digital geometry, images and video sequences, being funded to a total of 24,000,000 Chinese Yuan (about 1,600,000) by the Chinese Ministry of Science and Technology.The topics to be investigated will includeShape, image and video understanding, using neural networks and other ideas based on the structure of the human visual system, with an emphasis on high-performance systemsThe development of mathematical tools which are necessary for processing geometry and image and video understanding, allowing computers to learn about things they see, and to then classify new examples shown to them later. The key issue here is the huge amount of information present in the large numbers of pixels present in images, and even more so in video.How analogies to the human visual system can be used to help develop systems which can understand images and video. This includes such ideas as how to separate the big picture from the small details, using understanding at different geometric scales, and how to make methods which can automatically cluster related information. Further key issues are how geometric and visual information should be acquired and represented so that it can be processed effectively and efficiently to answer queries about content.How key features of geometric shapes, images and video can be represented, stored, compressed and encrypted. Particularly important are issues of digital rights management, including embedding invisible watermarks into visual data to enable proof of ownership, and the ability to provide content-aware security for geometric models, images and video.Other goals include new approaches to capturing and editing geometric models, images and video, including for example video blending.