Victor-Emil Neagoe, Serban-Vasile Carata
Subject Independent Drunkenness Detection Using Pulse-Coupled Neural Network Segmentation of Thermal Infrared Facial Imagery
Drunkenness detection, thermal imagery, image segmentation, pulse-coupled neural network (PCNN), genetic algorithms
This paper proposes a new model of subject-independent drunkenness detection based on analysis of thermal infrared facial images. The method consists of the following processing stages: (a) thermal infrared image acquisition; (b) Pulse-Coupled Neural Network (PCNN) image segmentation; (c) feature selection using Principal Component Analysis (PCA) cascaded with Linear Discriminant Analysis (LDA); (d) Support Vector Machine (SVM) classification. We have built an experimental thermal infrared facial image database of 10 subjects (7 males and 3 females). The thermal images of each subject have been acquired both for sober condition and also for inebriation condition obtained after the person drank a specific amount of alcohol. Any thermal picture has been taken using a FLIR camera and it corresponds to the resolution of 160 x120 pixels in the wave range of 7.5-13 µm. The parameters of the PCNN have been optimized using a genetic algorithm. Using the proposed thermal image analysis cascade based on PCNN, we have obtained a drunkenness detection score of 97.5%, corresponding to an increase of 17.5% over the best score given by the considered benchmark method without PCNN segmentation.
Cite this paper
Victor-Emil Neagoe, Serban-Vasile Carata. (2016) Subject Independent Drunkenness Detection Using Pulse-Coupled Neural Network Segmentation of Thermal Infrared Facial Imagery. Mathematical and Computational Methods, 1, 305-312