This talk will give brief overview of feature Detection in computer vision, which is essential step in almost every computer vision system. Reliable feature detector and descriptor facilitate succeeding computer vision tasks.  The talk will focus on methods based on local features, covering both intensity-based and structure based approaches. Several state of the art feature detectors and descriptors such as SIFT, SURF, Shape Context will be highlighted. 

"Towards Human Augmented Image Annotation (Image Annotation through Gaming)" by Lasantha Seneviratne

Recent developments in social networks have contributed to the already large quantity of digital multimedia content on the World Wide Web (WWW). As a consequence, the following questions arise, do people label the content? If so, how often do they do so? Reacting to these and other similar questions, researchers around the world have designed a considerable number of algorithms and frameworks with the capabilities of automated image tagging. However, due to the semantic gap, finding a general solution for image tagging has become a challenge. Addressing this issue, we designed an innovative approach to obtain an accurate label for images by taking into account the social aspects of human-based computation. The proposed approach is highly discriminative in comparison to an ordinary content-based image retrieval (CBIR) paradigm. It aims at what millions of individual gamers are enthusiastic to do, to enjoy themselves within a competitive environment. It is achieved by setting the focus of the system on aspects of the gaming environment, which involves human players. Furthermore, this framework integrates a number of different algorithms that are commonly found in image processing and game theoretic approaches to obtain an accurate label. As a result, the framework is able to assign (or derive) accurate tags for images by eliminating annotations made by a less-rational (cheater) player. The performance analysis of this framework has been evaluated with a group of 440 players. The result shows that the proposed approach is capable of obtaining a good annotation through a small number of game players.