LIP FEATURE EXTRACTION BASED ON PCA

摘要

The lip-reading has caused increasing focus of attention, in which feature extraction determines the recognition performance. Firstly, this paper employs the principle component analysis as the feature extraction approach. A mouth shape database is established, which consists of five Chinese vowels pronunciation lip images. Then, several experiments have been conducted to investigate the performances of PCA applied to lip feature extraction. The statistical results indicate that the average recognition accuracy of pronunciation mouth i and o is up to 70%, with maximum 100% recognition accuracy.