Chinese
English
Distinguished Research Fellow  |  Liao, Mark  
 
contact
vita
education
experience
interests
descriptions
activities
invited_talk
invited_visit
honors
publications
patents
software
supervised
 
 
 
 
 
Research Descriptions
 

        My current research interests include: developing video-based surveillance systems and developing core technologies for the processing and management of multimedia data collected by the National Digital Archives Program.

        The research on video-based surveillance system is concerned with the development of efficient human motion analysis algorithms and their applications to real-world problems, for example, the problem of "counting the number of people who are watching TV wall." In the past two years, our emphasis was put on developing an abnormal-action detection system and a trajectory-based event detection system. These results were published in IEEE Transactions on Multimedia , Vol. 9, No. 6, pp.1193-1201, Oct. 2007 and Vol. 10, No. 3, pp.372-384, April 2008, respectively. The core technology of the trajectory-based event detection system has been successfully transferred to a company.

        The technologies developed for the National Digital Archives Program (NDAP) can be divided into two main categories:(1) Smart Tone Reproduction technique for digital images/videos; (2) image/video inpainting techniques. The NDAP has collected a huge amount of digitized multimedia data in the past five years. These data include: images, audios, videos, texts, and 3-D graphics. Since part of these digitized data (such as digitized old pictures or old films) are with poor quality and they cannot be taken again, it is of crucial importance to develop new techniques to strengthen their quality. In 2007, we have developed a smart tone reproduction technique that is designed specifically for digitized images of the NDAP. The National Science Council of Taiwan has held a press conference in September 2007 to introduce this technique to the general public in Taiwan. At present, the developed technique is able to "recover," on average, 80-90% of the visual details of an image without doing any parameter adjustment.

 
 
bg