Face Recognition based on 2D-PCA and CNN hongliang xue
Motivation Face recognition technology is widely used in our lives Using MATLAB ORL database
Database The ORL Database of Faces taken between April 1992 and April 1994 at the Cambridge University Computer Laboratory 10 different images of each of 40 distinct subjects. http://www.cl.cam.ac.uk/research/dtg/attarchive/facedatabase.html
2D-PCA Conventional 1D-PCA: transform image matrix to 1D vector 2D-PCA: use image matrix to form a covariance matrix easier to determine corresponding eigenvectors
2D-PCA Train_num(per class) Test_num(per class) d Accuracy(%) Time(s) 3 7 8 85.36% 1.76 5 91.5% 1.806 9 1 95% 1.582 2 90% 1.525 16 92.5% 1.64 32 1.86
2D-PCA
CNN
CNN using deepLearnToolbox-master written by Rasmus Berg Palm https://github.com/rasmusbergpalm/DeepLearnToolbox still tuning parameters get about 15% error rate using model of 2 layers of convolution
Conclusion 2D-PCA: simple algorithm, accuracy quite high ( 90~95% ) may not perform well for larger dataset CNN: hard to find optimal parameters, takes a lot of time can perform well for large dataset
References 1. Jian Yang; Zhang, D.; Frangi, A.F.; Jing-Yu Yang, “Two-Dimensional PCA: A New Approach to Appearance-Based Face Representation and Recognition”, in Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 26, no. 1, pp. 131-137, January 2004 2. Lawrence, S.; Giles, C.L.; Tsoi, A.C.; Back, A.D. “Face Recognition: A Convolutional Neural-Network Approach”, in Neural Networks, IEEE Transactions on, vol. 8, no. 1, pp. 98-113, January 1997 3. http://www.cl.cam.ac.uk/research/dtg/attarchive/facedatabase.html 4. Lec 16 Deep Neural Network (2), Yu Hen Hu
Questions