Open AccessEnhanced Deep Learning with featured transfer learning in Identifying Disguised Faces
Author(s) -
Yedavalli Venkata Rao Raghava,
Venu Madhav Kuthadi,
S. Rajalakshmi
Publication year - 2019
Publication title -
international journal of innovative technology and exploring engineering
Language(s) - English
Resource type - Journals
ISSN - 2278-3075
DOI - 10.35940/ijitee.h7286.0881019
Subject(s) - computer science , artificial intelligence , convolutional neural network , deep learning , face (sociological concept) , task (project management) , facial recognition system , transfer of learning , pattern recognition (psychology) , face detection , convolution (computer science) , machine learning , artificial neural network , social science , management , sociology , economics
Paper The objective of face recognition is, given an image of a human face identify the class to which the face belongs to. Face classification is one of the useful task and can be used as a base for many real-time applications like authentication, tracking, fraud detection etc. Given a photo of a person, we humans can easily identify who the person is without any effort. But manual systems are biased and involves lot of effort and expensive. Automatic face recognition has been an important research topic due to its importance in real-time applications. The recent advance in GPU has taken many applications like image classification, hand written digit recognition and object recognition to the next level. According to the literature Deep CNN (Convolution neural network) features can effectively represent the image. In this paper we propose to use deep CNN based features for face recognition task. In this work we also investigate the effectiveness of different Deep CNN models for the task of face recognition. Initially facial features are extracted from pretrained CNN model like VGG16, VGG19, ResNet50 and Inception V3. Then a deep Neural network is used for the classification task. To show the effectiveness of the proposed model, ORL dataset is used for our experimental studies. Based on the experimental results we claim that deep CNN based features give better performance than existing hand crafted features. We also observe that the among all the pretrained CNN models we used, ResNet scores highest performance.