Expression Recognition based on Local Directional Pattern in the Eye Region and Artificial Neural Network

Abstract

Artificial face expression recognition is a research topic that was started ever since 90s. On this paper, writer will elaborate all the supporting algorithm used on the experiment, starting from eye-region detection using Haar cascade classifier (Haar) and Harris corres corner detection (Harris), up to expression recognition based on Local Directional Pattern (LDP) and Artificial Neural Network.(ANN)

Haar is a segmentation method based on Integral Image and classifying method named Adaptive Boosting. Haar is pretty accurate algorithm to segment an Image of face, from its background. However its performance to segment an eye-region could still be increased, hence the additional usage of Harris.

LDP Feature Extraction is an 8-directional-edge- response based feature extraction on each pixel. The final output of this process is a histogram of LDP-code on each area of the Input Image.

Artificial Neural Network is a supervised learning derived from standard linear perceptron. Especially on the usage of hidden layer, which allows the system to classify a dataset that wasn’t linearly separable. On this experiment, the training phase will utilize back- propagation algorithm.

Keywords: Face recognition, local transititional pattern, contrast feature extraction, multi-layer perceptron, supervised learning, backpropagation, Haar, Harris, corner detection, eye region detection.