Face recognition system based on feature extration

Automatic face recognition system has been facing problems in recognizing subjects of varying ages. Age invariant recognition has been of great use in tracking people database especially in public domain systems like driving license, passport, and criminal records etc., this paper deals with a forensic face recognition which is robust to changes in age, pose, expression and illumination. It has a pre-processing stage in which all the background information including the hairy parts in the face is removed by thresholding. After preprocessing the thresholded image is divided into macro blocks. The scale invariant feature points are extracted from all the blocks by means of Scale Invariant Feature Transform. These extracted feature points are further refined by Taylor transformation technique and dominant orientation is assigned to every feature point. The common features between the blocks are grouped by means of Feature Discriminate Analysis. These features are classified with the ferns from the database by means of Naive Bayes Classifier. This approach tends to give more robustness to pose and expression variations thereby improved accuracy of 60% in FG NET database compared to other probabilistic approaches with a trade off in processing time.