Face Recognition Using Curvelet and Waveatom Transform

ii iii iv ABSTRACT The field of digital image processing is continually evolving. Nowadays, there is a significant increase in the level of interest in image morphology, neural networks, full-color image processing, image data compression and image recognition. This work deals with image recognition with the application of face recognition. Some people think that face recognition is an easy task for computer system as for humans, but in reality most of the face recognition systems can't achieve a complete reliable performance because there are many factors affect on the process of recognition like: large variations in facial approach, head size and orientation, and change in environmental conditions, all these factors makes face recognition one of the fundamental problems in pattern analysis, other factors that impact the performance are the accuracy of face location stage and the number of actual face recognition techniques used in each system. So face recognition from still and video images is emerging as an active research area with numerous commercial and law enforcement application. This research identifies two techniques for face features extraction based on two different multiresolution analysis tools; the first called Curvelet transform while the second is waveatom transform. The resultant features are inputted to train via two famous classifiers; one of them is the artificial neural network (ANN) and the other is hidden Markov model (HMM). Experiments are carried out on two well-known datasets; AT&T dataset consists of 400 images corresponding to 40 people, and Essex Grimace dataset consists of 360 images corresponding to 18 people. Experimental results show the strength of both curvelets and waveatom features. On one hand, waveatom features obtained the highest accuracy rate of 99% and 100% with HMM classifier, and 98% and 100% with ANN classifier, for AT&T and Essex Grimace datasets, respectively. On the other hand, two levels Curvelet features achieved accuracy rate of 98% and 100% with HMM classifier, and 97% and 100% with ANN classifier, for AT&T and Essex Grimace datasets, respectively. A comparative study for waveatom with wavelet-based, curvelet-based, and traditional Principal Component Analysis (PCA) techniques is also presented. The proposed techniques supersede all of them. And shows the robustness of feature extraction methods used against included and occluded effects. Also, indicates the potential of HMM over ANN, as they are classifiers.