模式识别系列讲座
lecture series in pattern recognition
题 目（title）：multi-level and networked image representation
讲 座 人（speaker）：t.s. huang and guojun qi (university of illinois at urbana-champaign)
主 持 人 (chair) ：prof. chenglin liu
时 间 (time)：june 4, 2013(tuesday), pm 16:00
地 点 (venue)：no.1 conference room (3rd floor), intelligence building
报告摘要（abstract）：
in this talk, tsh shall first describe some recent research projects and ideas of his group related to the broad topic of image representation. at the visual feature level (low-level), we have recently developed hierarchical gaussianization (hg), a patch-based location-sentive representation, which is a kind of soft version of "bag of words". hg hs been applied to a number of visual recognition tasks (including face recognition and face verification) with great success. at the semantic level (high-level), we are exploring the use of ontology to help inference; i.e., to use the relationships between image/object classes/labels to increase recognition accuracy and even to come up with new paradigms of doing inference. the relationships of particular interest are: "is a subclass of", and "co-occurrence". we have developed several ways of taking advantage of these two relationships. finally, in many web-based applications, visual data (images and video) are often embedded in "heterogeneous networks" (hn), which may involve social media. hn have a wide range of applications. to name but two: personalized ranking and recommendation; collaborative trustworthy sensing (crowd sourcing and mining of trusted knowledge from social media).
in the second part of this talk, gjq will expound on "mining trusted knowledge".
承办单位：模式识别国家重点实验室 |
