How do ear recognition biometrics work?

In the era of face masks, ear recognition can provide a viable alternative to more commonly used facial recognition. The human ear remains relatively unchanged, making it suitable for security authentication applications. It can also be easier to extract information from since it is not subject to variations like facial expressions.

Ear recognition biometrics analyze unique structural characteristics such as the helix, antihelix, tragus, antitragus, lobe, and concha. Compared with other biometric identifiers, the ear’s structure is relatively simple. However, ears are distinctive enough to be identifiable between identical twins (Figure 1).

Figure 1. The outer structure of the human ear contains many features that can support accurate biometric differentiation, even between identical twins. (Image: National Center for Biotechnical Information)

Ears are just as reliable as fingerprints for identifying individuals. And ears don’t age like faces, except the lobe drops lower over time. Ears have mostly uniform color distribution, simplifying analysis. Ear features are large enough to be captured from a distance using a simple digital camera.

Ear recognition is like facial recognition and can be performed using deep learning algorithms. However, the simpler structure of the ear also enables the use of other classification technologies like discrete curvelet transform (DCT) and principal component analysis (PCA).

DCT is a fixed mathematical tool that decomposes images based on geometric features like curves and edges. The decomposition is performed at different scales and orientations and is designed to support a detailed analysis of structures within the image. In ear recognition, DCT can be used for feature extraction within an ML model. The curvelet coefficients derived using DCT are further analyzed, and predictions are derived using ML to arrive at a detailed classification for ear recognition.

Principal component analysis (PCA) is another ear recognition and classification tool. It can simplify a data set by reducing its dimensionality, which can be a useful preprocessing step for neural networks. PCA helps identify the significant features or principal components and can speed the training process when dealing with high-dimensional data sets like biometric information.

Studies have been performed comparing the results from a DCT and ML classification system, combining various levels of PCA analysis with CNNs, and comparing the performance of CNNs without PCA (Figure 2).

Figure 2. Ear classification can be performed using deep learning ML (left) or CNNs (right). (Image: SpringerLink)

According to the National Center for Biotechnology Information, an optimized deep CNN has achieved an ear recognition accuracy of 97.36%. The model requires relatively little memory, making it suitable for use on embedded and handheld devices. In addition, it can be expanded to support large-scale surveillance systems for environments like shopping malls, airline terminals, and railway stations.

Summary

Ear recognition can provide a valuable alternative to other biometric identification protocols, including fingerprints and facial recognition. This contactless technique can be performed at a distance, even if the subject is wearing a face mask. CNNs for ear recognition have been demonstrated to be highly accurate, even when used on resource-constrained devices.