Abstract
The principal intention of this paper is to study face recognition across age progression at two levels: feature extraction and classification. In other words, this work aims to prove the benefit of replacing the Softmax layer of the Deep-Convolutional Neural Networks (CNN) by Extreme Learning Machine (ELM) classifier based on deep features computed from fully-connected layer of pre-trained AlexNet CNN model, in a context of age-invariant face recognition. Experimental results indicate that the ELM classifier combined with feature extracted by the pre-trained AlexNet CNN model worked effectively for face recognition across age progression. As significant highest mean accuracy rates are always obtained using ELM classifier. These results are more significant, following a 95% confidence level hypothesis test.
Highlights
Age-invariant face recognition, as a focus topic of face recognition in uncontrolled environment, is a very useful technology which may be applied in large real-world applications in which age compensation is required, like criminal and missing children identification, and biometric security systems.Dealing with aging related variations is a challenging task because age related effects differ for different individuals and it is in combination with external factors, like health conditions and lifestyle, which have been shown to contribute to facial aging effects (Lanitis et al (2009)), elaborating age-invariant face recognition systems becomes a major necessity.Feature extraction is a key step in face recognition system
This work aims to prove the benefit of replacing the Softmax layer of the deep-convolutional neural networks (CNN) by extreme learning machine (ELM) classifier based on deep features computed from fully-connected layer of pre-trained AlexNet CNN model in a context of age-invariant face recognition
We perform an extensive comparative study in which we consider Softmax classifier, Support Vector Machine (SVM) classifier, Extreme Learning machine (ELM) classifier, and Kernel Extreme Learning Machine (KELM) classifier, based on deep features computed from fully-connected layers ‘Fc6’, ‘Fc7’ and ‘Fc8’of a pre-trained AlexNet CNN model
Summary
Feature extraction is a key step in face recognition system. It involves extracting information which best represent the image and which are invariant in context of face recognition across age progression. Prior to developing Deep-CNN, multiple manual age-invariant methods have been proposed for features extraction (Sungatullina et al (2013),(Bereta et al (2013)), and (Ling et al (2009)), that are computed from low level characteristics and statistical representation. Deep-CNN has become the most common method in use for automatic feature extraction (Agrawal et al (2019)), (Shakeel et al (2019)), and (Moustafa et al (2020)). Deep-CNN can be used in transfer learning with features from the pre-trained CNN model, in a classification task
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
