Enhanced Deep Learning Framework for Face Verification Across Age Progression

Abstract

Facial aging is a texture and shape variations that affect the human face as time progresses. Current face verification across age systems lack the required efficiency to recognize facial shape and texture variations at the same time while maintaining high accuracy, so the need was to create a powerful model that could identify these variations efficiently. Currents frameworks focus on using handcrafted techniques only, while others focus on the use of pre-trained models, so there is a need to develop an efficient model to extract shape and texture features in addition to taking advantage of the characteristics and strengths of handcrafted systems and pre-trained systems accordingly. The main objective of this research is to develop a model capable of extracting both shape and texture variations from the facial image, by fusing both shape and texture descriptors with pre-trained deep learning model to obtain better accuracy. Sequentially, a new model was developed from scratch using deep learning capable of extracting the variations that occur on the face. The research explores the use of a deeper convolutional neural network model from scratch, with Histogram of Oriented Gradients (HOG) descriptor to handle feature extraction and classification of two face images with the age gap. We studied the effect of fused GoogLeNet pre-trained convolution network model with Histogram Orientation Gradient (HOG) and Local Binary Pattern (LBP) feature descriptors at decision level through Majority Voting technique to achieve a good performance of our proposed system for face verification The experiments are based on the facial images collected from MORPH and FG-NET benchmarked datasets. Combining deep CNN with LBP seems to give minimum accuracy than combining it with both LBP and HOG. On the other hand, combining deep CNN architecture with HOG proved to give the highest accuracy value, which is 99.85%. Despite the FG-NET dataset contains fewer images, it appears that there is no improvement in the accuracy of the MORPH dataset. The future work is to implements a deeper pre-trained convolutional neural network models to make a comparison, also conduct a fusion of these models at decision level to improve accuracy.