Phase efficient neural network using Curvelet features for face recognition

This paper presents a novel scheme for face recognition application, by utilizing curved singularities obtained from curvelet transform and trained on phase efficient neural network. The phase efficient neural network is formed by processing the statistical descriptor and smooth coefficients of curvelet transform with neural network and then post-process with phase only correlation (POC). Neural network minimizes the search space of face subjects by yielding the response values to POC. The match/mismatch recognition accuracy is based upon the peak detection from the POC surface. The amalgamation of two recognition techniques on curvelet features have enabled us to look into the new dimension of not only improving the accuracy of neural network but also to decrease the computational and time cost of phase only correlation.