基于稀疏表示的手势识别方法

Gesture recognition method based on sparse representation

为提高传统手势识别方法在手势偏转情况下的识别率与识别速度,提出一种改进的SRC手势识别算法。采用YCrCb自适应阈值分割模型对原始图像进行肤色分割,利用主成分分析法(PCA)提取其主要特征,采用K-最近邻方法从样本集中选取K个样本构成新的超完备冗余字典,通过求解l1范数最小化的问题完成分类识别。实验结果表明,采用的KNN-SRC算法相对传统手势识别算法拥有更高的识别率,降低了SRC算法的时间复杂度,当K取值为20时识别速度为普通SRC算法的4倍。

To improve recognition rate and recognition speed under the condition of gesture deflection for which the conventional gesture recognition methods do not perform well,an improved SRC gesture recognition algorithm was proposed.YCrCb adaptive threshold segmentation model was adopted to do gesture segmentation,and gesture binary images were got.Main features of gestures were extracted using principal component analysis（PCA）.K samples were selected from the training sample dataset using K-nearest neighbor algorithm to form a new over-complete redundant dictionary.The test sample was classified successfully by solving l1-norm minimum problem.Experimental results show that compared with classical recognition algorithms,KNNSRC algorithm proposed gets higher recognition rate and reduces the computational complexity in SRC.Besides,when K=20,the recognition speed is 4times higher than SRC.