Using DLBP texture descriptors and SVM for Down syndrome recognition

Down syndrome, the most prevalent chromosome disorder in mankind, occurs approximately in one per thousand infants born per a year. Also, life expectancy of people suffering from this irregularity has increased from 25 to 59 in the last decades. Recognizing such patients in critical and high security places like security gates could assist responsible people to make proper decisions. This irregularity causes a private facial view which differentiates regular people from patients. In this study, we have proposed a novel framework, which uses first and second order directional derivative local binary pattern (LBP) histograms for texture description then applies the support vector machine for classification, in order to distinguish Down syndrome population from healthy one. We have investigated and compared two methods for texture description: one method utilizes only first order directional derivative LBP and the other benefits from both first and second order directional derivative LBPs. The histogram bins values obtained from the mentioned descriptors have been used for training the support vector machine to classify Down and not Down population. The proposed approach has been implemented using a custom database collected from free web resources. Experimental results show PPV, NPV, sensitivity and specificity factors equal to 92.35%, 96.50%, 96.66% and 92% in the best case, respectively.