Facial Recognition System Using Mixed Transform and Multilayer Sigmoid Neural Network Classifier

Abstract

Facial recognition systems are critical components in numerous applications. They are used, for example, to prevent retail crime, unlock phones, find missing persons, protect law enforcement, and aid forensic investigations. In such real-world applications, the identification of facial information must be both quick and exact. The purpose of this study is to improve both the accuracy and speed of facial recognition. The proposed system reduces overall computational complexity by using a few simple algorithms and transforms. The grayscaling algorithm enhances the image, and the salient features are extracted using a mix of two transform families: the two-dimensional discrete wavelet transform and the two-dimensional discrete cosine transform. This combination exploits the nonorthogonality of the coefficients in both domains to preserve the essential details and perceptual qualities of the original image. A multilayer sigmoid neural network is used for classification since the expensive training stage can be performed offline. The trained network, which uses efficient computations, can be embedded in an online system for rapid classification. The efficiency of the system is an attractive property when processing massive information datasets with limited resources. The recognition system is tested with four freely accessible datasets: the ORL, YALE, FERET-c, and FEI. A test set based on the combination of all datasets is also utilized to evaluate the system performance. Results show that despite the reduction in complexity, the system still maintains high recognition rates as compared to the popular existing methods.