【摘要】 提出了一种基于混合高斯模型的隐马尔科夫模型(GMM-HMM)与手势轮廓特征的单幅手势图像识别方法,该方法采用YCr Cb空间阈值处理对手势图像二值化处理,针对理想感兴趣区域提出了一种还原最上层轮廓的新型轮廓算法.将每类手势轮廓特征作为HMM的观察值分别训练对应手势的HMM参数,建立所有手势的HMM模型.分别用Viterbi算法计算测试集数据与每个模型的条件概率来获得识别结果.实验结果表明,该方法不仅对手势库内的特定人的静态手势识别具有较好的效果,且对提取的其他人的静态手势图像识别率也较高.更多还原

【Abstract】 Based on Gaussian- mixture Hidden Markov Model( GMM- HMM) and gesture contour feature,this paper proposed a method to recognize gesture from a single image. The YCr Cb color space threshold processing was adopted to get the binary image of the original RGB image,and then a new algorithm was presented to restore the most ideal contour. As an observation vector,it is used to train hidden Markov model parameters related to the corresponding gesture,so all the gestures’ HMM model can be established. The gesture can be recognized by the Viterbi algorithm through calculating the conditional probability which describes the relationship between test set and each gesture’s model. The experimental results showed that the method performed well for the specific static gesture images from the gesture library,and it was also effective for the images taken by ourselves.更多还原