Article Preview
Top1. Introduction
Ear biometrics has been emerging as a predominant area of research in the field of biometrics in recent years. Due to its several merits including consistent shape, universality, possessing rich discriminant features in small region, higher user acceptance and non-intrusiveness, it has used in many applications such as personal identification, device access control, forensics, head tracking, video surveillance, etc. (Emersic, Struc, & Peer, 2017). As ear shape possesses a rich detail features and stable structure over a long time, the ear biometrics ensures its popularity based on a number of publications in (Benzaoui, Hadid, & Boukrouche, 2014; Emersic et al., 2019; Hassaballah et al., 2019; Sarangi et al., 2018; Youbi, Boubchir, & Boukrouche, 2018) as unimodal and (Banerjee, & Chatterjee, 2017; Hezil, & Boukrouche, 2017; Huang et al., 2013; Sarangi, Mishra, &, Dehuri, 2018; Toygar, Alqaralleh, & Afaneh, 2018) as multimodal biometrics. Despite of several improvements in ear recognition techniques, the improper ear localization could severely degrade recognition performance of the biometric system. Ear localization is the most challenging task of ear recognition systems. To date many related works have been proposed in the literature to develop an efficient ear localization technique. However, these existing techniques could not avoid the effect of all uncontrolled issues such as illumination change, skin color variation, lighting conditions, different head positions, scale variation, partial occlusion, ear accessories, and cluttered background. In addition, most of them are not fully automatic techniques and the processing time is also large that restricts them to be employed in real-time applications. As a consequence, to manage all those uncontrolled challenges, the ear localization is still an open research problem for the biometric community.
Over the last decades, swarm intelligence algorithm has been considered as a competitive approach for resolving numerous biometrics and machine learning applications (Razmjooy et al., 2013; Guo et al., 2019; Namadchian et al., 2016; Razmjooy et al., 2016). They are population based search algorithms which has potential to obtain global optimal solution in less iteration. In recent years the Jaya algorithm (Rao, 2016) has become popular because of its simple design and algorithmic specific parameter free evolutionary algorithm. Hence, Jaya algorithm can become much more suitable in real-world applications. Hence, we propose entropy-based binary Jaya algorithm (EBJA) to search dense edge regions of ear template size as a set of probable ear candidates from the skin regions in less computational time.
Furthermore, this paper intends to adopt the principle of Hausdorff distance measures that computes the similarity measure between two binary images using pixels coordinates rather than pixels intensity values. Recently, several researches have been made by using different variants of Hausdorff distance measures for face biometrics (Huttenlocher et al., 1993; Dubuisson, & Jain, 1994; Barnabas et al., 1998; Lin et al., 2003). The classical Hausdorff distance (CHD) in (Huttenlocher et al. 1993) and its variants are utilized for comparing binary images with an advantage that it is more robust to the small change of points location. Hence, it measures proximity instead of exact superposition. But, the CHD measure is very much sensitive to noise which affects the performance of object matching in the image. In (Dubuisson, & Jain, 1994), the proposed modified version of CHD measure referred to as modified Hausdorff distance (MHD) to suppress the effect of noise points and they showed better performance in presence of illumination variation, noise, and partial occlusion. In (Barnabas et al., 1998), further improved the MHD by introducing robustness to change in illumination and tolerance in non-rigid local distortions and named it as doubly modified Hausdorff distance (M2HD). Similarly in (Lin et al., 2003), authors introduced spatially Eigen-weighted Hausdorff distance (SEWHD) and spatially Eigen-weighted doubly Hausdorff distance (SEW2HD) in computing similarity measure among different face images, but both methods performance are greatly affected when changes of illumination in the neighborhood parts of the face are frequently occurred.