Article Preview
TopBackground
Biometrics has been broadly employed in many access control systems, digital security, and forensics. This is the technology that automatically recognizes a person by using the physiological or behavioral characteristics (Jain et al., 2004). These characteristics such as face, ear, fingerprint, iris, voice, and deoxyribonucleic acid (DNA), are collected by the instruments such as digital camera and voice recorder. The collected data are labelled, trained, and applied to identify individual among others. There is a myriad of biometrics which have been applied to our day-to-day life such as face recognition, DNA analysis, fingerprint identification, signature verification (Qi & Hunt 1994) and speaker recognition, etc.
Biometric has certain advantages that make it popular in recent years. Unlike passwords, biometrics is much difficult to be shared and guessed (Jain, 2007). For an example, passwords combined with numbers or alphabetical letters normally contain personal information such as birth date or name. Complicated ones are hard to be remembered and easy to forget. On the other hand, it is tough to replicate the biometrics information of a person, and biometrics does not need to be remembered.
Nowadays, face recognition is one of the most commonly used biometrics (Turk & Pentland, 1991), however, it may not be employed in twin recognition directly. The obvious problem of twin recognition is uniqueness of the faces (Gauthier & Logothetis, 2000). Twin recognition using face recognition is one of the most difficult tasks (Mahalingam & Ricanek, 2013) which has greatly challenged the conventional face recognition. Recently, many researchers have paid their special attention to this problem and have obtained notable achievements.
Research of Biswas et al. (Biswas, Bowyer & Flynn, 2011) shows that human visual system can work much better than computational machines in twin recognition. The research outcomes find that human is good at acquiring facial marks such as moles and scars to effectively identify twins. In addition, if human beings can spend more time to carefully examine the difference, the performance will become much more reliable. Klare, Paulino and Jain (2011) identify twins with human facial features such as eigenfaces as the level 1, Local Binary Patterns (LBP) as the level 2 and facial marks as level 3. The mouth shapes are relatively helpful to separate twin in typical data sets; especial the focus is on the age changing of twins (Le, Luu, Seshadri & Savvides, 2012). Therefore, they have employed facial aging features in twin identification. In another literature, Vijayan et al. (2001) utilize a 3D model to recognize twins. Albeit a spate of good recognition has been achieved, it is still far from solving the problem finally, e.g. closing the gap to human visual system.
Figure 1(a) are photos of identical twins taken when they were two months old. It is really hard to distinguish them from these photos. Figure 1(b) was the time when they were between two and ten years old. The two photos shown in Figure 1(c) were acquired when the twins were between 15 and 23 years old. In these three groups of photos, twin A is at the left side and twin B is at the right side. The twins look very similar in all the photos indeed.
Figure 1. Growth of two twin brothers: (a) Twins of months old (Left A Right B); (b) Twins of 2, 6 and 10 years old; (c) Twins of 15, 18 and 23 years old (ageing)
TopThere are a number of methods in face recognition of twins. Dimensionality-reduction approach is the primary mission in these algorithms. The methods usually take the entire face for consideration (Delac, Grgic & Grgic, 2005). Principal Component Analysis (PCA) which adopts the feature eigenface is one of the effective algorithms in face recognition (Gottumukkal & Asari, 2004; Moghaddam & Pentland, 1997). PCA greatly reduces the computational complexity by diminishing dimension of features.