VGGFace-Ear: An Extended Dataset for Unconstrained Ear Recognition.

Area of Science:

• Biometrics and Pattern Recognition
• Computer Vision
• Artificial Intelligence

Background:

• Ear biometrics offer unique advantages for identification, including contactless capture and distance accessibility, making them suitable for surveillance and security.
• Traditional biometric methods often require large, labeled datasets for training, which are scarce for ear images captured in uncontrolled environments.
• Convolutional Neural Networks (CNNs) are powerful tools for image recognition but are data-hungry, posing a challenge for ear recognition tasks.

Purpose of the Study:

• To develop and introduce a new, large-scale dataset of ear images captured under uncontrolled conditions, exhibiting significant inter-class and intra-class variability.
• To evaluate the effectiveness of transfer learning using pre-trained CNNs for ear recognition on unconstrained datasets.
• To establish a benchmark for future research in unconstrained ear biometrics.

Main Methods:

• A novel ear image dataset was constructed by leveraging the existing VGGFace dataset, comprising over 3.3 million images.
• Transfer learning was employed, utilizing CNN models pre-trained on large-scale image and face recognition tasks.
• Two distinct experiments were conducted on separate unconstrained ear image datasets to assess recognition performance.

Main Results:

• The developed ear image dataset demonstrates high variability, reflecting real-world conditions.
• Transfer learning with pre-trained CNNs achieved promising results in ear recognition tasks on unconstrained datasets.
• Rank-based metrics were utilized to report performance, providing a quantitative evaluation of the proposed method.

Conclusions:

• The newly created ear image dataset is a valuable resource for advancing research in unconstrained ear biometrics.
• Transfer learning offers an effective approach to overcome the data scarcity issue in ear recognition using CNNs.
• The findings support the potential of ear recognition as a viable biometric modality for security and surveillance applications.