LBP-based ear recognition

Abstract

The ear, as a biometric, has been given less attention, compared to other biometrics such as fingerprint, face and iris. Since it is a relatively new biometric, no commercial applications involving ear recognition are available. Intensive research in this field is thus required to determine the feasibility of this biometric. In medical field, especially in case of accidents and death, where face of patients cannot be recognized, the use of ear can be helpful. In this work, yet another method of recognizing people through their ears is presented. Local Binary Patterns (LBP) is used as features and the results are compared with that of Principal Components Analysis (PCA). LBP has a high discriminative power, tolerance against global illumination changes and low computational load. Experiments were done on the Indian Institute of Technology (IIT) Delhi ear image database and results show that LBP yields a recognition rate of 93 % while PCA gives only 85 %.

Learning pairwise SVM on hierarchical deep features for ear recognition

“…In this work, we use deep CNN for ear feature extraction. Instead of deep features, many hand-crafted features, including both holistic [27] and local [7,[28][29][30] ones, have been suggested for ear images. Compared with hand-crafted features, deep CNN features are learned in a data-driven manner, and have proven its effectiveness in many vision tasks.…”

Metric Learning with Dynamically Generated Pairwise Constraints for Ear Recognition

“…For the appearance-based features, it mainly includes intensity, directional and spatial-temporal information. Many of works have presented holistical features such as Eigenear and Eigenface [2], ICA [13], active shape model to detect outer ear contour [14], 1 D and 2 D Gabor filter [15,16], and locale features such as LBP [17], HOG [18], kernel of polar sine transform (PST) [19], SIFT [20], and SURF [21]. Recently, Nanni and Lumini [22] adopted the sequential forward floating selection (SFFS) to select the best features from sub-windows in an ear image.…”

Image-based Ear Biometric Smartphone App for Patient Identification in Field Settings

“…2D approaches are more appropriate for our domain because of the field requirements of fast and cheap solutions. Extracting a feature vector from a 2D ear representation has been done in many ways including Eigen Ears (PCA) [Chang et al, 2003], Force Field [Abdel- Mottaleb and Zhou, 2005], GFD [Abate et al, 2006], SIFT/SURF [Cummings et al, 2010, Kisku et al, 2009, and LBPs [Wang et al, 2008, Boodoo-Jahangeer andBaichoo, 2013].…”

“…LBPs [Ojala et al, 1996] demonstrated good performance in ear biometrics [Wang et al, 2008, Boodoo-Jahangeer andBaichoo, 2013]. Following [Takala et al, 2005, Wang et al, 2008 we use P sampling points on a circular grid of radius R, and uniform binary patterns u2 originally introduced by [Mäenpää et al, 2000] (LBP u2 (P,R) ).…”