Investigation of effectiveness of ensemble features for visual lip reading

Features used for classification play essential role in the performance of system. In the field of Lip reading, features appear in large number which has to be solved by selection of subset of features. Work covered in this paper validates the performance of individual visual features such as lip height, lip width, area of lip region, angles at corners and then combine them to create a new subset feature that improves the classification accuracy of certain weak features when combined with significant attributes. Each feature provides different level of representing classification characteristics for words. Area feature provided highest independent accuracy of 75.70%. Ensemble feature area-h4 produced highest combined accuracy of 71.25%.