Hand gesture recognition based on improved histograms of oriented gradients

To reduce the influence of edge information in background, an improved feature extraction method for hand gestures in which the histograms of oriented gradients is combined with the skin similarity is proposed. Weight computed on the skin similarity is introduced into the gradient of each image pixel. This new gradients can enhance the hand features. Histograms of oriented gradients with different size of cells are employed to classify the hand gestures because different sizes of cells depict different local features. The experiment results indicate that the size of the cell affects the recognition rate greatly and the combination of histograms of oriented gradients with two appropriate different cells can represent the hand gesture features well. Images with different time, illumination and background collected by real environment and images in Marcel database are recognized by the improved method and the results show that the proposed method can improve the hand gesture recognition.