Article Preview
Top1. Introduction
Biometrics is defined as the automation of human recognition frameworks using distinguishing traits. Among those traits, face recognition is still being as one of the most challenging research topics dealing with data analysis for pattern recognition and the understanding of interactions between humans and machines. This is mainly due to the presence of confounding facial differences such as lighting, expression and such structural components as hair, beard, glasses and so on (Jain, 2008). The features of the human face can be represented by pattern vectors, which can then be viewed as data points distributed in a high-dimensional feature space. However, such encodings are unstable, and it is computationally time consuming to conduct face recognition procedures directly in such a high-dimensional feature space. Although many efforts and some progress have been made in this field, accurately recognizing the face under realistic condition settings is still difficult. In this regard, Various methods have been proposed to address all these drawbacks and problems in the face recognition system (Zhao, 2003). For example, bi-dimensional subspace techniques such as 2DPCA (Yang 2004) and 2DLDA(Li, M 2005) combine sub space reduction and features extraction outcomes, and it is able to obtain spatial discriminant feature from the original space, which is extremely useful for image classification task. The techniques have drawn tremendous interest from both academic and industrial biometric areas and has been developed rapidly during the past decade. The process generally involves image projection into a low-dimension space for feature extraction and classification task. Reducing the redundant information of the image is the key for building effective biometric systems. However, a subspace projection approach’s is facing the problems of small sample size or high dimensional data and in face classification and recognition tasks. These are known to provide enhanced recognition rates relative to 1D PCA methods (Oravec, 2004), but it is possible that performance can be further improved by first stabilizing the images. One technique which can achieve this is the Iterated Function System (IFS); this has been widely used as an effective feature extraction and reduction technique in other fields of research such as data compression (Barcellos. 2014; Jacquin, 1989; Fisher, 2012). IFS also has been used in several biometric applications such as iris recognition (Teo, Chuan Chin. 2005), gait recognition (Zhao, 2006), and face recognition (Ebrahimpour-Komleh, 2001; Mohamed. 2015).). However, the main difficulty with IFS is the computational complexity involved in the Partitioned Iterated Function System (PIFS) process. Several other trials of IFS in face recognition have been proposed. For example, it has been used to extract features from the human face (Ebrahimpour-Komleh. 2001, Wen-Shiung 2013). In addition, a novel fractal feature extraction algorithm optimized by genetic algorithm for human face recognition has been developed (Jemaa Y, 2011). In addition, Temdee et al. propose the fractal dimension (FD) method and neural network-based architecture for face recognition (Temdee, 1999). A fast fractal coding method has been applied for face recognition and carried out on three public data bases (Tang, Z, 2018). While IFS displays a performance accuracy comparable or higher than other techniques, a number of features, in particular computational time and parameter stability, make it impracticable as a stabilization technique in realistic situations. However, can be adjusted to satisfy the requirements of effectively biometric system.