A Comparative Analysis Performance of Data Augmentation on Age-Invariant Face Recognition Using Pretrained Residual Neural Network
No Thumbnail Available
Date
2021
Journal Title
Journal ISSN
Volume Title
Publisher
Journal of Theoretical and Applied Information Technology
Abstract
There has been an immense improvement in face recognition research. Unfortunately, the accuracy of face recognition systems recognizing the same person over time due to ageing is open research. Minor geometric changes in the face that occur due to ageing contribute to face recognition systems' inaccuracy. Researchers,
over subsequent years, have come up with methods to improve the performance of Age Invariant Face Recognition (AIFR) systems, the most recent one being the use of Convolutional Neural Network (CNN) to create face recognition models. The pre-trained residual network (ResNet) is trained and tested using a heterogeneous database to actualize this improvement. The heterogeneous database consists of images from 82 Caucasian subjects in the FG-Net database and 11 African subjects. These obtained images were augmented using geometric transformation and Noise to increase the amount of data for training. Afterwards, a model robust is developed. The Sliding Window framework was used to detect the faces fed into CNN for training and testing. After getting the results from our classification model, an analysis was carried out on the classification models of both the original dataset and augmented datasets. It was observed that the model performed remarkably with the noise-injected dataset and performed worst with the geometric transformation database.
Description
Journal Article
Keywords
Citation
Abere, R. et. al.(2021) A Comparative Analysis Performance of Data Augmentation on Age-Invariant Face Recognition Using Pretrained Residual Neural Network; Journal of Theoretical and Applied Information Technology: Vol.99. No 6. pp1309 - 1319. https://www.jatit.org/volumes/Vol99No6/6Vol99No6.pdf