Influence of the Intra-Modal Facial Information for an Identification Approach

Influence of the Intra-Modal Facial Information for an Identification Approach

Carlos M. Travieso (University of Las Palmas de Gran Canaria, Spain), Marcos del Pozo-Baños (University of Las Palmas de Gran Canaria, Spain), Jaime R. Ticay-Rivas (University of Las Palmas de Gran Canaria, Spain) and Jesús B. Alonso (University of Las Palmas de Gran Canaria, Spain)
DOI: 10.4018/978-1-4666-5808-0.ch013
OnDemand PDF Download:
$37.50

Abstract

This chapter presents a comprehensive study on the influence of the intra-modal facial information for an identification approach. It was developed and implemented a biometric identification system by merging different intra-multimodal facial features: mouth, eyes, and nose. The Principal Component Analysis, Independent Component Analysis, and Discrete Cosine Transform were used as feature extractors. Support Vector Machines were implemented as classifier systems. The recognition rates obtained by multimodal fusion of three facial features has reached values above 97% in each of the databases used, confirming that the system is adaptive to images from different sources, sizes, lighting conditions, etc. Even though a good response has been shown when the three facial traits were merged, an acceptable performance has been shown when merging only two facial features. Therefore, the system is robust against problems in one isolate sensor or occlusion in any biometric trait. In this case, the success rate achieved was over 92%.
Chapter Preview
Top

Introduction

In this chapter, we develop a new theory in which the features individually and separate human beings can provide at least the same information embedded in the face. In other words, the whole is not greater than the sum of the parts.

Figure 1.

Different facial features

In this chapter, we explore the possibility of applying the model introduced in this new psychological theory, in order to improve the performance of traditional face recognition systems. This theory highlights the importance of face recognition based on the analysis of the features and adding the results of the analysis of the whole face. That is, to build a facial identification system based on the model of the human recognition according to the latest findings.

The objective of this chapter is to develop and implement a biometric identification system by merging different intra-multimodal facial features, such as the mouth, eyes and nose. Thereby, the main aim is to improve the efficiency of identification systems based on a single feature for the identification and provide alternatives to the existing fusion methods.

Furthermore, this chapter try to reinforce the hypothesis relate to the face recognition process from the physiological human point of view. Therefore, it will be proven that the sum of the analysis of each trait separately and subsequently, the merge the data obtained, provides at least the same information as the whole face when the system is configured in identification mode for identify subjects Additionally, the isolated traits and the whole face biometric were combined in order to analyze if this combination improve the performance of the system. Finally, it was intended to test the efficacy of multimodal systems against possible occlusions or problems in any uni-modal sensor, highlighting the advantages of the multi-biometric systems.

In order to develop a robust and reliable system, has been found the best combination and configuration of pre-processing tools, which require less computation time and offer the highest success rate. The Independent Component Analysis (ICA), Principal Component Analysis (PCA) and Discrete Cosine Transform (DCT) were used as parameterization techniques. It was established a comparison that allow select the one that best suits to each trait. The classification system is implemented by using Support Vector Machines (SVM).

The following figure shows the general scheme of the system implementation.

The identification and verification of people has been a goal to accomplish by humans throughout history. In fact, in most processes involving people, his identity is almost as important as the actions carried out. For example, in the communication process, the sender, the receiver or the transmitter of the message are, in most cases, as important as the message.

Furthermore, the identification of people is especially critical in security restricted or forensic identifications, in which an identification error can have serious consequences to property or persons. Therefore, techniques have been developed that allow the recognition from differentiated characteristics of each person such as voice, face, fingerprint, iris, signature verification, hand geometry, etc. These characteristics are called biometric modalities (Bolle et al., 2003).

From these technologies can derive various applications. For example, in the areas of physical and logical access control to keep information or properties, in airports, healthcare system, financial system, vehicle environment, working desk, personal computer and also in the area of control for the presence attendance tracking and work timetable. They can also be useful in the field of justice and public order to facilitate administration and revenue in prisons, identification at crime scenes and other forensic applications.

The goal of the developers of facial identification systems is to find the system that matches or even improve the ability of humans to recognize faces. Therefore, it is interesting to ask how the system identification of human faces and if possible or beneficial resemble the systems developed human model. This is an issue that has been carried out a long time and, although there are several theories, has not reached a conclusion that allows say categorically how to identify human faces.

Complete Chapter List

Search this Book:
Reset