In this paper, we propose an improved multimodal ear and profile face biometrics for human identity recognition under uncontrolled conditions such as illumination changes, pose variation, low contrast, partial occlusion, and blur. In this framework, ear and profile face images are localized from the side face image which is acquired using a single sensor. The feature vectors for both the images are separately extracted using Gabor wavelets to minimize the effect of image degradation. Here, kernel canonical correlation analysis (KCCA) is exploited for feature-level fusion over canonical correlation analysis (CCA) that outperforms in generating discriminant feature vector. Finally, the nearest-neighbor classifier is applied for classification of personal identity. The proposed multimodal biometrics is evaluated on two public databases, namely, University of Notre Dame collection E and J2. Experimental results show that KCCA-based fusion improves both identification and verification performance over CCA and yielding promising results compared to other existing unimodal and multimodal biometrics.
CITATION STYLE
Sarangi, P. P., & Panda, M. (2021). Combining Human Ear and Profile Face Biometrics for Identity Recognition. In Lecture Notes in Electrical Engineering (Vol. 744 LNEE, pp. 13–24). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-33-6781-4_2
Mendeley helps you to discover research relevant for your work.