CiiT International Journal of Digital Image Processing, Vol 7, No 08, August 2015 234 0974-9691/CIIT–IJ-6055/08/$20/$100 © 2015 CiiT Published by the Coimbatore Institute of Information Technology Abstract---Lip password is composed of a password embedded with motions of lip and point out the characteristic of lip motion. To provides security of a speaker verification system by using private password and behavioral biometrics of a lip motion simultaneously. The target speaker saying wrong password then rejected and the target speaker saying correct password then detected. Here a Hidden Markov Model (HMM) learning approach based on multi boosted scheme is presented for a security speaker system. This method first extract the visual features and to characterize each frame. The lip password segmentation algorithm is used for the segmentation of lip sequences. Hidden Markov Models with boosting learning framework contains random subspace method and data sharing scheme. Finally, the lip-password is verified based on verification results provided by all the subunit learned from HMM based multi-boosted scheme and it will check whether the password is spoken by the speaker with the already-recorded password or not. Keywords---Lip Motion, HMM, GMM, RSM, DSS I. INTRODUCTION OWDAYS in the community considerable attention must needed for speaker verification because of its several applications such as secure access control, financial transaction authentication, human-computer interfaces, security protection [1], [3]. This verification process is done by following. Using previously stored information first it verifies a claimed speaker under a certain criterion; thereby the speaker would be either accepted or rejected based on verification [1]. Traditionally, not only the linguistic information is conveyed by speech also it characterizes the speaker identity that can be utilized for speaker verification. To achieve speaker verification the most natural modality may probably be acoustic speech signals. Speaker verification system based on acoustic model is effective in domain of application, but the performance of the purely acoustic SPV would be degraded dramatically because of the environment corrupted by the multiple talkers or background noise. For instance, Kenny et al. derives the spatial and temporal models Manuscript received on July 28, 2015, review completed on July 28, 2015 and revised on February 08, 2015. A.Jebaselvi is with the Department of Information Technolgy in Manonmaniam Sundaranar University, Tirunelveli, TamilNadu, India. E-Mail: jebaselvia21@gmail.com. Dr. Kumar Parasuraman is with the Centre for Information Technology and Engineering, Manonmaniam Sundaranar University, Tirunelveli, Tamilnadu, India. E-Mail: kumarcite@msuniv.ac.in / kumarcite@gmail.com Dr. T. Arumuga Maria Devi is with the Centre for Information Technology and Engineering, Manonmaniam Sundaranar University, Tirunelveli, Tamilnadu, India. E-Mail: deviececit@gmail.com. Digital Object Identifier: DIP082015003. for the identification of speaker by extracting the group of visual lip region features and then utilizes the Hidden Markov Model with Gaussians Mixtures [10],[15],[16], while Sao et al. Gaussian Mixture Model are utilized to build the statistical speaker models. Later, Shafait et al. utilized the GMM classifier and extracted a suitable visual features group from the sequential mouth regions. Faraj et al. Utilizes GMM and lip motion group features are extracted with person verification. The lip motion alone concentrated on this paper, although the other modalities can also be fused with this. The password is attached with the movement of lip and the basic characteristic of motion of lip sequences is composed to derive a concept of lip motion password. This approach provides a double security lip-password protected speaker verification system [9]. Simultaneously, the password information and behavioral biometrics of lip motions are verified by the speaker. Whether the wrong password said by the target speaker or the accurate password known by wrong person will be discovered and refused accordingly. The speaker verification system is efficient means it has some advantages and they are as follows. Motions of lip sequences are not sensitive to the environmental noise; the lip motion acquisition is not affected by distance; this type of speaker verification system can be carried out in a hidden way silently; simply applicable to a weakened speech person. In literature, almost all related speaker verification systems analyze the whole utterance as the basic unit of processing for the lip motions. The lip password protected system is designed based on the capability that it simultaneously detects both of the following two cases. That is the speaker saying incorrect password and wrong person saying the accurate correct password. Such tasks are incompetent in these traditional methods. Generally, the lip-password protection system has always comprises of many sub block units. A short period of lip motions are indicated by these subunits and to describe the underlying lip-password information these subunits always have diverse moving between various elements that are considered individually, but not as a whole. Here, we present a multi-boosted Hidden Markov Model learning approach to such a speaker verification system based on lip-password. A group of representative visual features are extracted first to characterize each lip frame and then an effective algorithm is used to segment the lip-password sequence. Then these sequences of lip-password are splitted into a small set of sub block units. For subunit verification we integrate Hidden Markov Models with multi boosted learning Security Based Speaker Verification for Lip-Password using Learning Multi-Boosted HMMS A. Jebaselvi, Kumar Parasuraman and T. Arumuga Maria Devi N