Feature extraction technique using hybridization of DWT and DCT for gender classification

In this paper, a robust technique to construct feature vector for gender classification has been proposed. Discrete Wavelet transform is used in concatenation with Discrete Cosine transform to form the feature vector. Initially, multi-level Discrete Wavelet transform is applied to images to obtain the approximation coefficients of image. Discrete Cosine transform are then calculated for the obtained approximate image. Hybridisation of DWT and DCT reduces the feature vector size significantly. Using this feature vector as input, SVM classifies the images. 2-Fold cross validation dataset is used to learn the SVM optimal parameter. Face images of three different databases i.e. AT@T, Faces94 and Georgia Tech databases are used to evaluate the efficiency of proposed technique for gender classification. Results show that the proposed technique performs better as compare to other state-of-art techniques.