The implementation of real-time person authentication system using visual lip feature

Abstract

It is the human nature that each individual does possess some kinds of unique biological and/or behavioral characteristics which are generally referred to as biometric characteristics. Due to uniqueness, biometrics can be used to recognize the identity or to verify the claimed identity. Authentication is a process to match the claimed person’s identity with the biometric information stored in the accessing system and the candidate will either be accepted or rejected depending on the matching scores. This topic has become a focus in past decades. Recently, automatic speech recognition using bimodal information involving audio and visual cues has aroused the interest of many researchers since lip information can characterize person identity. Nevertheless, solely using lip features for authentication is a relatively new research area and only limited articles can be found in the literatures. This thesis presents the studies of the appropriateness of various kinds of lip features for person authentication. A real-time authentication system solely based on the lip information has been developed. Lip segmentation, lip modeling, lip feature extraction and authentication performance are major issues of the system and they will be addressed in this thesis. In order to obtain the lip features, the first step is to segment the lip region from the skin/background region. The fuzzy c-means (FCM) clustering based algorithm has been shown to be effective and reliable for this task even for lip image with low color contrast. By integrating the spatial distance information to the clustering objective function, the lip region can be accurately segmented. A 14-point Active Shape Model (ASM) is used to model the outer lip contour. With this ASM lip model and the lip membership map produced by segmentation algorithm, a probability based cost function can be established for defining the lip region and in turn the lip contour. The ASM lip model parameters can be obtained with the use of a fast convergent point-driven optimization technique developed earlier in our group, a robust lip contour initialization procedure has been developed to improve the system efficiency for processing image sequence. With the ASM lip model, the geometric dimension and the shape related feature parameters can easily be computed. In this study, we have evaluated the use of geometric parameters, ASM shape-based parameters, inner lip features, lip intensity profile and their derivatives for authentication. These parameters are trained with the Hidden Markov Model (HMM) to produce a reference template for storing in the accessing system. A person can claim his identify to the authentication system and then utter a password in front of a camera, without audio sound for secrecy, and the system can make the decision by comparing the features against the stored template of the claimed person. Experimental results have shown that the ASM shape based parameters, the inner lip features and the intensity profile produce the best results. A real-time authentication system using the proposed combination of lip features has been implemented and running on a 2.4 GHz PC. The field trial experiment demonstrates encouraging results and confirms that visual lip features are an alternative potential biometrics for authentication applications.

Date of Award	15 Jul 2005
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Wing Hong Ricky LAU (Supervisor)