A lightweight convolutional neural network built on inceptio-residual and reduction modules for deep facial recognition in realistic conditions

ABSTRACT

Since the last two decades, face recognition has been a significant area in the fields of biometrics and machine learning that uses the different appearances and geometric-based facial features to recognize the living substances present in a frame or video. In this article, we have proposed an inceptio-residual convolutional neural network inspired by the ResNet architecture for deep facial feature extraction and recognition. Besides that, some of the well-known lite or moderately dense and highly accurate ImageNet benchmarked convolutional neural networks, such as VGG-19 BN, ResNet 18/34/50, SE-ResNet50, ResNeXt50, Inception V2/V3/V4, Xception50, and DualPathNet68, have been utilized for face recognition. In conjunction with that, the proposed one was tested along those above-standard networks with respect to various performance metrics. Based on the results, we concluded that the proposed Incepto-residual convolutional neural network has performed better than the ImageNet benchmarked Convolutional Neural Networks.

KEYWORDS:

• Face recognition
• ResNet
• inception module
• residual function
• Inceptio-residual module
• reduction module
• Inceptio-Reisdual convolutional neural network
• Data Augmentation

Acknowledgement

We are grateful to the Georgia Institute of Technology's Center for Signal and Image Processing, the California Institute of Technology’s CalTech, Cambridge University's Vision and Robotics Group, Yale University's Computational Vision and Control Center, the Department of Electrical Engineering, and the Department of Electrical and Computer Engineering at The Ohio State University for making their standardized face recognition datasets (Georgia Tech, CalTech Faces 1999, AT&T, Yale, and CFEE, respectively) freely available.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Ashutosh Satapathy

Ashutosh Satapathy, Assistant Professor, is with the Department of Computer Science and Engineering, Velagapudi Ramakrishna Siddhartha Engineering College, Vijayawada, AP 520007, India. (e-mail: [email protected]). He earned his Ph.D. in the School of Computer Science and Engineering at VIT, Chennai. He received the ITRI international internship during his Ph.D. study in 2017. Previously, he had worked as a lecturer at the Odisha University of Technology and Research (formerly the College of Engineering and Technology, Bhubaneswar). His areas of interest include image processing, high-dimensional data analysis, and pattern recognition, and he has published 11 articles in various international journals and conferences.

Jenila Livingston L. M.

Dr. Jenila Livingston L. M., Professor, is with School of Computer Science and Engineering, VIT, Chennai, TN 600127, INDIA. (e-mail: [email protected]). She has completed her Ph.D. in Faculty of Engineering from National Institute of Technical Teachers’ Training and Research (NITTTR), Government of India, Chennai and master’s degree in Computer Science and Engineering from Anna University, India. She has nearly 19 years of experience in Teaching and Research and keenly interested in the areas of eLearning, Engineering Education, Artificial Intelligence, Soft Computing, Data Analytics, Internet & Web Programming, Data Base Systems, and Data Structures & Algorithms.