【作者】 路文焕； 曲悦欣； 杨亚龙； 王建荣； 党建武；

【Author】 LU Wenhuan;QU Yuexin;YANG Yalong;WANG Jianrong;DANG Jianwu;School of Computer Software, Tianjin University;School of Computer Science and Technology, Tianjin University;

【机构】 天津大学软件学院； 天津大学计算机科学与技术学院；

【摘要】 在基于超声的无声语音接口实现中,通常使用主成分分析或离散余弦变换提取舌部超声图像的特征。为了保留图像的关键信息,本文提出了三种混合特征提取方法:第一种方法称为Wavelet PCA,使用主成分分析从小波系数中提取特征;第二种和第三种方法分别称为block DCT-PCA和block WHT-PCA,首先根据能量选取适量的离散余弦变换或沃尔什-哈达玛变换系数,然后使用主成分分析提取选定系数的特征。实验结果表明,本文提出的混合特征提取方法优于主成分分析或离散余弦变换,其中block DCT-PCA方法具有最好结果。更多还原

【Abstract】 Principal Component Analysis(PCA) and Discrete Cosine Transform(DCT) are used to extract the feature of ultrasound image to build the ultrasound based Silent Speech Interface(SSI). In order to preserve the critical information of image, this paper presents three hybrid feature extraction methods. The first method is wavelet PCA, which uses PCA to extract Discrete Wavelet Transform(DWT) coefficients feature. The second and third methods are called block DCT-PCA and block WHT-PCA, respectively, in which the DCT or Walsh-Hadamard Transform(WHT) coefficients are truncated to appropriate number according to energy, and then the truncated coefficients are used by PCA to extract feature. The experimental results show that the proposed hybrid feature extraction methods outperform the standalone PCA or DCT. And the block DCT-PCA method gains the best result among all the methods.更多还原