ABSTRACT
There has been a significant amount of research in recognizing humans through images of the iris captured under near-infrared illumination and constrained scenarios, whereas recognition under visual spectrum and unconstrained scenarios is relatively recent and challenging. Hence, several attempts have been made by researchers to identify humans not only through the iris but also through recognizing patterns existent in the periocular (periphery of the ocular) region. In such biometric systems, images of the periocular region are considered as biometric templates to be used
Abstract 153 6.1 Introduction 154 6.2 Literature Review 157 6.3 Why Localization of the Periocular Region Is Required 159 6.4 Proposed Periocular Localization Methods 161
6.4.1 Through Human Anthropometry 161 6.4.1.1 Detection of Sclera Region and Noise
Removal 165 6.4.1.2 Content Retrieval of Sclera Region 166
6.4.2 Through Demand of Accuracy of Biometric System 169 6.4.3 Through the Subdivision Approach and
Automation of Human Expertise 171 6.5 Conclusions 171 References 172
for recognition. Effort has been made in this article to specify a rectangular boundary around the eye region (periocular region) in a facial image that is potentially optimal for human recognition. A comparatively larger template of the periocular image can be slightly more potent for recognition but slows down the biometric system by making feature extraction computationally intensive and increasing the database’s size. A smaller template, however, cannot yield desirable accurate recognition, although it performs faster because of the low computation needed for feature extraction. These two contradictory objectives (i.e., to minimize the size of the periocular template to be considered for recognition and to maximize recognition through the template) are intended to be optimized through the proposed research. This article proposes four different approaches for dynamic optimal localization of the periocular region. Feature extractors are found to work efficiently on the periocular region when employed to evaluate the proposed localization. The optimal localization methods are tested on the publicly available unconstrained UBIRIS.v2 and FERET databases and satisfactory results have been achieved. The results assure that optimization of the aforementioned two objectives can be achieved to mark an optimized size of the periocular region.
