基于SVD的唇动视觉语音特征提取技术

唇动视觉语音特征提取是音视频驱动的人脸动画唇动表示和唇读研究的关键技术．首先针对彩色视频图像进行唇色增强，对增强后的灰度图像进行闽值分割，获取唇部包围框，并根据口型发音的视觉特征进行初分类；然后进行尺度与灰度归一化处理，对预处理后的图像提取奇异值特征；最后采用基于欧氏距离的模板匹配法对该奇异值特征所包含的视觉语音信息进行测试试验．结果表明该低维度特征包含了大量唇动视觉语音信息，可用于单个人在自然环境下的唇语口型识别．

Visual-audio feature extraction of lip movements based on SVD

Visual feature extraction of lip movement is a key issue in video and speech driven face (animation systems. )The approach is like this: firstly enhance chromatic video image, then segment enhanced gray images with thresholds, and finally obtain lip shapes. This method classifies the lip-shapes according to the visual features of pronunciations, regulates the dimensions and grayscales of lip images, and (extracts) features based on SVD from the preprocessed images. Finally, the template (matching algorithm based ) on Euclidean distance is applied. The results show that the character of lower dimensions includes a large number of visual speeches' information, so it can be applied in individual natural conditions.