Threshold-based outer lip segmentation using support vector regression

Automated lip reading from videos requires segmentation. Threshold-based segmentation is straightforward, but it rarely used. This study proposes a histogram threshold based on the feedback of shape information. Both good and bad examples were used to train an $$\epsilon $$ -support vector regression model infer accuracy region shape. The was optimised minimise error. proposed method tested 895 images 112 subjects using AR Face Database. method, implemented in simple algorithms, reduced errors by 23.1%.