Convolutional Deep Learning Framework for Age Invariant Face Recognition

Abstract

Face recognition is diversely used in security scenarios such as surveillance, forensics, automatic photo annotation, fighting child trafficking and most importantly identification of criminals. Most of the face recognition techniques developed in the recent past tend to show good results in a constrained environment. However, numerous unforeseen problems arise while employing these techniques in real-world scenarios such as low quality face images, temporal variations and intentional or un-intentional face disguise creating variations in facial features. Additionally, the varying facial appearance with the passing time results in introduction of significant intraclass variations, which further makes face recognition a challenging task In this thesis, a deep learning framework has been proposed by exploring the efficacy of deep learning methods incorporating simple noise-based data augmentation for age invariant face recognition (AIFR). The proposed method detects face in an image using Viola Jones face detector and classifies it using a pre-trained Convolutional Neural Network (CNN) fine-tuned for AIFR. During transfer learning, a pre-trained CNN learns generalized age invariant features from facial images of several subjects to correctly identify them under varying age period and facial disguises. We have compared the performance of nine pre-trained 2D CNNs, for AIFR, each with a different number of learning parameters, based on their classification accuracy and execution time for selection of an accurate as well as a fast and efficient model. Experiments have been carried out on face and gesture recognition network ageing dataset (FG-NET-AD) for AIFR. The promising results achieved in these experiments reflect the efficiency of our proposed model. In the detailed comparison carried out with modern AIFR techniques, our proposed model outperforms them in all aspects. The research work carried out in this thesis has been further enhanced to include disguise invariance alongwith AIFR. For this purpose further detailed experiments have been carried out on six challenging disguised faces databases alongwith performance comparision of four pretrained CNNs for selecting a model that is suitable for both AIFR and Disguise invariant face recognition (DIFR).