User Independent Static Hand Gesture Recognition System using Boundary Information

摘要

Hand gesture can be object of analysis in many scientific fields due to some of its intrinsic characteristics. Specifically in pattern recognition, it can be studied as a problem that presents a large number of classes and also high variation among the patterns within a class. To deal with such characteristics, we propose to use MLP neural network. The features, based on the boundary information, are presented as histograms to attenuate the effect of non-linear boundary deformations. The performance was evaluated using a 26 postures database, captured from different users. The method presented better performance when compared with DP Matching. In addition, the number of classes in the system can also be extended without significant growth in the processing time.