Wavelet Domain Image Symmetry Detection Applied To Face Location

Symmetry detection is important in the fields of computer vision and image processing that interested many researchers. Wavelet transform is a new signal analysis theory developed in past two decades. Image symmetry detection in wavelet domain and its application to facial location are investigated in this thesis. The essential contents can be listed as follows:A method to implement multi-scale symmetry transform in the wavelet coefficients domain decomposed using anti-symmetrical biorthogonal wavelets (ASBW) is presented with application to location of feature points on human face image. Compared to the generalized symmetry transform, which is performed on a single resolution level, our method can yield results with higher location accuracy and less computational burden.A novel approach to reflectional symmetry detection with application to face detection is given using multi-scale gradient angle obtained from ASBW decomposition coefficients. The method is robustness to different illuminant conditions. Experiments with MIT database show that this method can yield results with higher detection accuracy and decreasing computational cost compared to the similar method.In addition, a method to image reflectional symmetry detection and normalization based on principal component analysis(PCA) is presented. Experiment results with facial images are given.