T.T.T. Bui, N.H. Phan, V.G. Spitsyn

A novel algorithm using wavelet transform and principal component analysis method is proposed for face and hand gesture recognition on digital images. The proposed algorithm testing results are presented. It is shown that use of the proposed algorithm gives effective performance of face recognition and hand gesture recognition on digital images. In this paper, an original complex algorithm based on Viola-Jones method, wavelet transform and principal component analysis method is also proposed for multiple face recognition on video. The examples of multiple face recognition on video sequence are resulted. It is shown that use of the proposed algorithm allows recognizing multiple faces on video sequence in real time. In this paper a part of face database Collection of Facial Images, containing 336 persons and 20 images of each person (summary 20×366=7320 images), is created for testing face recognition algorithm. In this paper two face database are created for training this algorithm. The first one contains 5 images of each person (summary 5×366=1830 images), and the second one - 10 images of each person (summary 10×366=3660 images). Testing results of the proposed algorithm show that the number of rightly recognized faces is 98,4 %. In order to test the performance of hand gesture recognition algorithm, in this paper a part of Cambridge Gesture database is used. This testing database consists of 5 parts. The contrast condition of each part is not the same. All the hand gestures are divided into 12 classes using for recognition. For each part one hand gesture database, containing 200 images of each class (summary 12×200=2400 images), is created for testing this algorithm. The combining testing database of all 5 parts contains 1000 images of each class (summary 12×1000=12000 images). For each part one training database also is created. This training database contains 20 images of each class (summary 12×20=240 images). The combining testing database of all 5 parts contains 100 images of each class (summary 12×100=1200 images). Testing results of the proposed algorithm show that the number of rightly recognized hand gestures is 94,63%.