【作者】 赵向阳； 张有为；

【Author】 Zhao Xiangyang Zhang Youwei (The Institute of Information Science,Wuyi University,JiangmenCity,Guangdong,China 529020)

【机构】 五邑大学信息科学研究所；

【摘要】 本文提出一种视觉语音识别中的函数变形模板灰度轮廓向量表征法,它是基于传统变形模板和灰度轮廓模型,采用函数变形模板自动求导数和边沿点垂线,自动训练建立了嘴唇灰度轮廓模型.与传统的边沿梯度相比,更好地表征了嘴唇轮廓特征,进而构造了更简单、更接近于二次型的能量函数,选择了更合适的多变量寻优算法,使嘴形特征提取的准确性、鲁棒性都有一定提高,更接近于实用化。实验表明唇动序列跟踪准确性提高了13.8%.更多还原

【Abstract】 This paper proposes vector expressing method of function deformable template with intensity profile in visual speech recognition.It bases on the conventional deformable template method and intensity profile model.It trains and sets up a intensity profile model automatically by taking advantage of the character of the function deformable template which can gain derivative and perpendicular of each edge dot automatically.Compared with conventional edge gradient expressing,the model stands for the character of mouth profile better.Then,a simpler and more approximative quadratic form energy function is constructed,and one more proper ndimesional line minizations is selected,so that the accuracy and robust of extracting mouth features are improved greatly,which makes it easier to be put into practice.Experimental result on the accuracy of tracing lip moving improves by 13.8%.更多还原