The article presented an HMM-based mm lip recognition for limited users of a handheld device. Tests are made on two small databases. The balanced accuracy in case of three, five, and ten classes are observed. While accuracy in case of three classes (two users) is approximately 99%, it falls to approximately 90% when ten classes (nine users) are considered. From the confusion matrices, it is evident that this fall of accuracy of 10% is due to the increase in number of classes. As this research focuses on use of handheld device by limited number of users, this limitation of scalability is not an issue. This approach can satisfactorily produce performance in considered situation. For practically using this methodology, any template replacement algorithm can be embedded into the biometric system to overcome slight challenges faced due to seasonal change of lip.

Chapter Contents:

• 6.1 Introduction
• 6.1.1 Evolution of lip as biometric trait
• 6.1.2 Why lip among other biometric traits?
• 6.1.3 Biometric authentication for handheld devices
• 6.1.4 Suitability of lip biometric for handheld devices
• 6.2 Motivation
• 6.3 Anatomy of lip biometric system
• 6.3.1 HMM-based modelling
• 6.3.2 Training, testing, and inferences through HMM
• 6.4 Experimental verification and results
• 6.4.1 Assumptions and constraints in the experiment
• 6.4.2 Databases used
• 6.4.3 Parameters of evaluation
• 6.4.4 Results and analysis
• 6.5 Conclusions
• References

Inspec keywords: image fusion; hidden Markov models; image texture; feature extraction; image recognition; shape recognition; matrix algebra; biometrics (access control); mobile computing

Other keywords: HMM; confusion matrices; hidden Markov model; mobile devices; lip biometry; lip recognition; texture features fusion; shape features fusion; handheld device

Subjects: Computer vision and image processing techniques; Mobile, ubiquitous and pervasive computing; Sensor fusion; Image recognition; Algebra; Algebra; Markov processes; Markov processes