Face Recognition Using Hybrid Techniques

Abstract

In real life applications whether it is government use (Law enforcement, Security/Counterterrorism, Immigration or Voter verification) or commercial use (Residential security, banking using ATM or physical access of buildings areas, doors or cars) identification of individual subject has become necessary for authentication purpose. Most of the techniques have been used from authentication perspectives like password based authentication or pin code based authentication for a subject in banking devices and smart phones. Some drawbacks also appeared for using these techniques as there was a possibility of password or pin codes to be stolen or hacked so, from security point of view these techniques might not be as good as it would have to be. Another technique named Finger print authentication has been used by security agencies and in the telecom sector as well, where thumb mark of an individual is essential and also for biometric verification. Another technique for identification named as Face Recognition has shown its role in areas of research and from the implementation point of view. In this technique face of an unknown subject is being first detected and then recognized after classification and then by comparison of it with other faces being stored in database. A Face image is full of information but using all the information will be time consuming and less efficient, so there is a need to use some of features of faces for identification and recognition purposes. This method has found its implementation in law enforcement agencies. For pattern recognition perspective the feature extraction is the most important task where we are required to obtain the best matching features with low processing time and for the best results. Another perspective to have concern is to reduce the dimensionality and complexity in the process that may also lead to reduce the time to process. In research perspectives several techniques has been proposed by different people in this method in different era for better recognition results. In this research work some existing techniques are merged to get the recognition results by using the methods of Principal component analysis (PCA) and SVD along with Hidden Markov Model (HMM) also the mean square error and PSNR are calculated.