Estimation of Geometrical Distortions of Lip Contours in Visual Input Systems
Abstract
Introduction: Lip reading is a method of extracting speech data from video information. The existing lip-image speech recognition systems are sufficiently powerful when the speaker is located en face in front of the camera, i.e. when the images are observed with only linear distortions, not causing any change in the shape of the lips. When the speaker is located at an angle, nonlinear distortions come out, and the shape of the lips changes. The problems of compensating the nonlinear distortions have almost never been discussed. Purpose: The goal is to develop an algorithm for estimating the geometric distortions of the lip contours which would make it possible to improve the existing systems of retrieving voice information from video data. Results: A technique has been proposed for estimating geometrical distortions of lip contours in visual information input systems. The geometric distortion parameter is estimated by the results of calculating the normalized scalar product of the observed contour of the speaker’s lips and the transformed contours, as the distortion parameter is changing. Practical relevance: The proposed algorithm for the estimation of the type and parameter of a contour distortion allows you to promote the efficiency of recognizing distorted contours of lip images in systems of visual input of speech information from video data.Published
2017-08-21
How to Cite
Khafizov, R., & Yaranceva, T. (2017). Estimation of Geometrical Distortions of Lip Contours in Visual Input Systems. Information and Control Systems, (4), 2-6. https://doi.org/10.15217/issn1684-8853.2017.4.2
Issue
Section
Information processing and control