Because of lacking full hand gestures contour information,current commonly used hand gesture recognition algorithms using single feature have lower recognition rate for the gestures with high local similarity and complicated shapes.Therefore we proposed a novel hand gesture feature extraction method,which combines the feature descriptor of curvature scale space (CSS )with Hu invariant moment. First,we used the skin colour model to extract the gestures from complicated background,and then extracted Hu invariant moment and CSS descriptor of gestures respectively to construct fusion features.At last,we made use of the artificial neural network to recognise and classify the new features.Experimental results demonstrated that compared with the recognition approaches based on single gesture feature,the proposed method has higher integral recognition rate,and improves significantly in recognition rate on gestures with high local similarity in shape.%目前常用单特征手势识别方法中，缺少完整的手势轮廓信息，对局部相似度高和形状复杂的手势识别率较低，为此提出一种将CSS特征描述子与Hu不变矩相结合的手势特征提取方法。首先，利用肤色模型把手势从复杂的背景中提取出来，然后分别提取手势的Hu不变矩和CSS描述子来构建融合特征，最后利用人工神经网络对新特征进行识别和分类。实验结果表明，与基于单一特征的识别方法相比，该方法整体识别率更高，对局部形似度高的手势识别率有很大提升。
展开▼
