Face recognition is used in many applications, such as security systems, card verification, video surveillance, credit criminal identification, person identification [2], passport renewal, law enforcement and biometric authentication [3].
[2] M. Sharif, F. Naz, M. Yasmin, M. A. Shahid and A. Rehman," Face Recognition: A Survey," Journal of Engineering Science & Technology Review, vol. 10, no. 2, 2017. [3] M. M. Sawant and K. M. Bhurchandi, "Age Invariant Face Recognition: A Survey on Facial Aging Databases, Techniques and Effect of Aging," Artificial Intelligence Review, pp. 1-28, 2018.
Face recognition is considered an effective and applicable identification and verification technique, as it is an inexpensive and non-intrusive approach [1].
In these cases, aging leads to change the important features of the face image. Hence, face recognition across aging can be considered as a common issue in many security and surveillance systems.
Age-invariant face recognition
Age-invariant face recognition is the task of performing face recognition that is invariant to differences in age.
In the field of face recognition, there are predominantly two problems: the recognition of images of the same person at different ages, and the estimation of a person’s age through a test image.
Typically, they are addressed through three methods: Discriminative, Generative, and Deep Neural Networks (DNN). For all three methods, the models are trained to learn (or derive) key features that are invariant over time or aging. These are then used for the identification of test or query images. \cite{Mittal2023}
In the discriminative methods, the focus is on using the invariant features to classify test images.
In the generative methods, the invariant features are used to generate images at various ages, and then these are used to match with the test images.
Generally, discriminative methods have been used for classification and generative methods for age estimation.
Deep neural network based methods are used for both classification and age estimation purposes.
Human face images consist of many pixels. If the size of the image is nm, then the dimension of the face is n*m; it is approximately 100 or 1000 dimensions or many more, depending on the size of the face image. So face image data sets are considered a high dimensional dataset. The face images with variability like illumination, pose, expression, and age gap are not linearly separable. The principal component analysis (PCA) technique does not perform well on this type of variable face dataset for classification/recognition types of problems, as PCA linearly separates the data point and is a linear classifier. Therefore face images are non-linearly separable.
As described above, face images lie on high dimensions and are non-linearly separable. Therefore, it becomes a difficult task to classify the face images. The main focus of any classification problem is extracting the discriminative features from the training datasets, which a classifier uses for recognition and verification purposes.
AIFR has four major frontiers, the face aging datasets, which have face images that consist of variations in many dimensions along with aging,