lip region segmentation and feature extraction method for Speech Recognition

摘要

PURPOSE: A method for detecting a lip region for speech recognition and acquiring lip shape information using the lip region is provided to detect a face region and a lip position using a pattern classifier employing a face color and lip color distribution model learning result and a support vector machine. CONSTITUTION: When an exemplary face image is input and a lip region on the face image is designated(S501,S503), a color distribution histogram of the lower part of the face image is calculated and a color distribution histogram of the designated lip region is calculated(S504,S505). Lip color probability for each color is obtained using the two color distribution histograms(S507). The lip color probability constructs a lip color similarity table(S509). A pattern classifier for lip shapes is trained using various examples of lip shapes(S510).