الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 115 |  |  |
| | | from | 115 | | |
Abstract
Lip movement is a visual information source tightly and synchronously coupled to the acoustic speech act and hence can be viewed as an integral part of a speech recognition effort. Most approaches to automated speech recognition that consider solely acoustic information are very sensitive to background noise or fail totally when two or more voices are present simultaneously, as often happens in offices, conference rooms, outdoors, and other real-world environments. There are different techniques used to extract the visual features of the speech from the image of the speaker lip, such as active contour model ”snake” and principal components analysis ”PCA”. The modified snake model was proposed to extract the lip contour on every video frame. Because the contour feature should be consistent and comparable among all video frames, snake contour was selected to be the feature, which is described using m radial vectors. The radial vectors are uniformly spread in 3600 and each of them originates from the centroid of the snake contour. The snake algorithm looks for contour features in the geometric space. Also principal component analysis ”PCA” has been successfully used for feature extraction of mouth images in the eigenspace. PCA reduces the high dimensional image data to low dimensional feature vectors without losing relevant information. PCA was applied to extract the principal components in the eigenspace, then the most significant eigenvectors were chosen, which used to project a mouth image from the huge image space to the eigenspace with much lower dimensions. The projected weights on eigenvectors define a feature vector.