Face Detection using DT-CWT on SHPCA Space

A novel face detection algorithm is presented by applying dual tree complex wavelets transform (DT-CWT) on spectral histogram PCA space (SHPCA) and support vector machine (SVM). DT-CWT is a transform recently studied, which provides good directional selectivity in six different fixed orientations at different scales. It has limited redundancy for images and is much faster than Gabor transform to compute. Hence, DT-CWT is a good choice to replace Gabor transform in some image signals processing fields especially for face images representation. In the face detection algorithm presented in this paper, images are first projected to SHPCA space after convolved with different filters including DT-CWT filters to achieve features subtraction based on frequency. Then on SHPCA space SVM classification is applied to detect whether faces exist in images or not. The experimental results show that DT-CWT performs much better than Gabor transform on SHPCA space. Furthermore, during preliminary experiments, SVM based on SHPCA space has been trained on a training set of 4000 faces aligned and 6000 non-face images, and a robust classifying function for face and non-face pattern is obtained, which gives the satisfying performance. Several questions about computation time saving and performance improvement are discussed