Gesture Recognition Based on Fusion Features from Multiple Spiking Neural Networks

Gesture recognition is one of challenging image processing. In this paper, a method of gesture segmentation is proposed, which is based on fusion of multi-information from multiple neural networks inspired by the human visual system. In this method, the gesture region is segmented from the video image sequence, innovatively using integration of the outputs from two kinds of spiking neural networks. The structures and the properties of the two networks are detailed in this paper. Based on the integrated outputs, the features of distance distribution histograms and outline moments are extracted and fused to form the mixed features. Finally, gestures are classified by the multi-class Support Vector Machine. Experimental results show that the proposed algorithm works efficiently and can perform gesture segmentation and gesture recognition with the satisfying accuracy for dynamic visual image sequence under complex background. It is promising to apply this approach to video processing domain and robotic visual systems.