Visual media include information such as images, video, and digital geometry. It is unstructured, the amount of data is large, and it comes in a variety of semantic categories. Nowadays, visual media computing provides a common base for the digital content industry, public security, visual media online services and other nationally significant demands. For example, images and video data account for 90% of internet traffic, and there are thousands of surveillance cameras collecting a tremendous amount of video and images in China. Therefore, effective and efficient acquisition and computational content analysis techniques are crucial issues in visual media computing.
Professor Hu Shimin and his research group from the Department of Computer Science and Technology at Tsinghua University have been recognized for their work on the project “Theory and Methods for Geometric Computing in Visual Media”. Starting with the geometric properties of visual media, they showed how geometric descriptions and analysis based on inner geometric structure and invariant mechanisms could contribute to effective acquisition and efficient algorithms for content analysis. The results achieved by their project on “Theory and Methods for Geometric Computing in Visual Media” won the second prize in the 2015 National Natural Science Award of China. The main author is Porfessor Hu Shimin from Tsinghua University, while the other project members include Professor Huang Jiwu from Sun Yat-Sen University, and Professor Ai Haizhou, Associate Professor Xu Kun and Dr. Chen Tao from Tsinghua University.
They proposed a new computing approach relying on intrinsic geometric structures, enabling them to break through the limitations of traditional methods that lack structured and parameterized representations. They introduced computational intelligence corresponding to visual cognition, and formed theories and methods for geometric computing in visual media. These methods are able to deal with many difficult problems such as visual media retrieval, recognition, composition and copyright protection.
The publications in this project (20 papers) have been cited 3,116 times in total; they were cited by SCI 804 times; 8 representative papers were cited 1,847 times, and by SCI, 510 times. Citations came from MIT, Stanford and 220 other universities/institutes; they were cited by 27 ACM/IEEE fellows and were highly praised by the academic and the industry community. Many international media have reported the achievements of this project. Some are already incorporated into university textbooks and teaching material, and are having a widespread international influence.