Journal of Information Organization Abdul Manan Ahmad, Sami M Halawani 3 3 2013 http://www.dline.info/jio/fulltext/v3n3/1.pdf There are many approaches to detect hand gestures. To narrow the search region, most existing methods construct fixed postures. However, under natural condition it is not realistic to make a user stick on a certain pose. In this paper a method to quickly detect hand shape quickly and robustly is experimented. Firstly, a skin color detector will be utilized to detect the presence of a hand in the image, then a set of image is clustered using k-mediod algorithm. Next, a tree structure of boosted cascades will be constructed. A general hand detector is provided by the main tree while the branches will classify valid shapes. Our experiments show a detection rate of 80% and from that 85% is achieved for recognition.