Face recognition from a side profile view, has recently received significant attention in the literature. Even though current face recognition systems have reached a certain level of maturity at angles up to 30 degrees, their success is still limited with side profile angles. This paper presents an efficient technique for the fusion of face profile and ear biometrics. We propose to use a Block-based Local Binary Pattern (LBP) to generate the features for recognition from face profile images and ear images. These feature distributions are then fused at the score level using simple mean rule. Experimental results show that the proposed multimodal system can achieve 97:98% recognition performance, compared to unimodal biometrics of face profile 96.76%, and unimodal biometrics of ear 96.95%, details in the Experimental Results Section. Comparisons with other multimodal systems used in the literature, like Principal Component Analysis (PCA), Full-space Linear Discriminant Analysis (FSLDA) and Kernel Fisher discriminant analysis (KFDA), are presented in the Experimental Results Section.
Ear Recognition has recently received significant attention in the literature. Even though current ear recognition systems have reached a certain level of maturity, their success is still limited. This paper presents an efficient complete ear-based biometric system that can process five frames/sec; Hence it can be used for surveillance applications. The ear detection is achieved using Haar features arranged in a cascaded Adaboost classifier. The feature extraction is based on dividing the ear image into several blocks from which Local Binary Pattern feature distributions are extracted. These feature distributions are then fused at the feature level to represent the original ear texture in the classification stage. The contribution of this paper is three fold: (i) Applying a new technique for ear feature extraction, and studying various optimization parameters for that technique; (ii) Presenting a practical ear recognition system and a detailed analysis about error propagation in that system; (iii) Studying the occlusion effect of several ear parts. Detailed experiments show that the proposed ear recognition system achieved better performance (94:34%) compared to other shape-based systems as Scale-invariant feature transform (67:92%). The proposed approach can also handle efficiently hair occlusion. Experimental results show that the proposed system can achieve about (78%) rank-1 identification, even in presence of 60% occlusion.
In this paper the problem of human ear detection in the thermal infrared (IR) spectrum is studied in order to illustrate
the advantages and limitations of the most important steps of ear-based biometrics that can operate in day and night time
environments. The main contributions of this work are two-fold: First, a dual-band database is assembled that consists
of visible and thermal profile face images. The thermal data was collected using a high definition middle-wave infrared
(3-5 microns) camera that is capable of acquiring thermal imprints of human skin. Second, a fully automated, thermal
imaging based ear detection method is developed for real-time segmentation of human ears in either day or night time
environments. The proposed method is based on Haar features forming a cascaded AdaBoost classifier (our modified
version of the original Viola-Jones approach1 that was designed to be applied mainly in visible band images). The main
advantage of the proposed method, applied on our profile face image data set collected in the thermal-band, is that it is
designed to reduce the learning time required by the original Viola-Jones method from several weeks to several hours.
Unlike other approaches reported in the literature, which have been tested but not designed to operate in the thermal band,
our method yields a high detection accuracy that reaches ~ 91.5%. Further analysis on our data set yielded that: (a)
photometric normalization techniques do not directly improve ear detection performance. However, when using a certain
photometric normalization technique (CLAHE) on falsely detected images, the detection rate improved by ~ 4%; (b) the
high detection accuracy of our method did not degrade when we lowered down the original spatial resolution of thermal ear
images. For example, even after using one third of the original spatial resolution (i.e. ~ 20% of the original computational
time) of the thermal profile face images, the high ear detection accuracy of our method remained unaffected. This resulted
also in speeding up the detection time of an ear image from 265 to 17 milliseconds per image. To the best of our knowledge
this is the first time that the problem of human ear detection in the thermal band is being investigated in the open literature.