SBB based complex background recognition system on hand gesture input

Weak Hypotheses learning has been applied wider and wider in machine learning. This paper proposes a novel multi-class SBB (Stress Based Boosting) algorithm based on the feedback mechanism. In real-time, recorded English sign letters are classified and recognized, and translated to Chinese characters finally. The system makes the Chinese HCI by hand gestures. At first, pre-processing is used to deal with the image sequence to eliminate noises and the background to reach the hand area and fixed-scale RoI detector. Then, with the help of feature detectors, feature points are extracted. The final analysis and classification proceed by the SBB algorithm which is aided by feedback factors to correct the error decisions on hand samples. In every period, feature learning is added. The algorithm can decide the classification strategy by concentrating the gesture training set from the results of classifications. The system is featured by self-learning and fast and accurate classification. According to the experiment results, the system has strong robustness, high recognition rates and good prospects for future application.