Face Detection using Min-Max Features Enhanced with Locally Linear Embedding

Face detection is critical function in many embedded applications such as computer vision and security as it is widely used as preprocessor for face recognition systems. As a preprocessor, the face detection system needs to extract features from a region of interest and classify them quickly as either face or non-face. In our previous works, we have devised a feature representation method called Min-Max (MMX) feature that allows representation of a region of interest using a few data points based on the unique characteristics of vertical and horizontal summation of face regions. In this paper, we attempt to improve the classification accuracy of MMX by integrating a technique called Locally Linear Embedding (LLE), a powerful dimensionality and feature enhancement algorithm that has been used successfully in many pattern recognition tasks. To test the performance of the proposed enhancement, the LLE-treated features were compared with non-treated features using a Multi-Layer Perceptron (MLP) neural network classifier. The results indicate an increase (+1.2%) in classification accuracy of the MLPs, demonstrating the ability of LLE to enhance the representation of MMX features.