Transfer Learning based Face Recognition using Deep Learning
M. Madhu Latha1, K. V. Krishnam Raju2
1M. Madhu Latha, Department of Computer Science and Engineering, SRKR Engineering College, Bhimavaram (Andhra Pradesh), India.
2Dr. K. V. Krishnam Raju, Department of Computer Science and Engineering, SRKR Engineering College, Bhimavaram (Andhra Pradesh), India.
Manuscript received on 11 May 2019 | Revised Manuscript received on 05 June 2019 | Manuscript Published on 15 June 2019 | PP: 38-44 | Volume-8 Issue-1S3 June 2019 | Retrieval Number: A10080681S319/2019©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Face Recognition technology is advancing rapidly with the current developments in computer vision. Face Recognition can be done either in a single image or from a tracked video. The task of face recognition can be improved with the recent surge of interest in the deep learning architectures. In this paper we proposed an Inception-V3 CNN using an approach called Transfer learning. Deep learning algorithms are trained to solve specific tasks and designed to work in isolated. The idea of transfer learning is to overcome isolated learning and leveraging knowledge acquired from one task to solve the related ones. Thus, instead of training CNN from scratch we use Google’s pretrained Inception v3 model by applying transfer learning. Inception v3 model was trained on ImageNet dataset and transferred its knowledge to LFW dataset to perform our task face recognition. LFW dataset is a collection of 13000 images of 5749 individuals. Using all the classes result in less accuracy because some of the classes may have single images, so we limited the dataset to 10 individuals with at least 50 images in each class to train the model. We also used a own dataset of 10 individual classes with 100 images for each class.
Keywords: Face Recognition, Deep Learning, Artificial Intelligence, Transfer Learning and Inception v3.
Scope of the Article: Deep Learning