Face Morphing Attack Detection in the Presence of Post-processed Image Sources Using Neighborhood Component Analysis and Decision Tree Classifier

Recently, Face Morphing Attack Detection (MAD) has gained a great deal of attention as criminals have started to use freely and easily available digital manipulation techniques to combine two or more subject facial images to create a new facial image that can be viewed as an accurate image of any of the individual images that constitute it. Some of these morphing tools create morphed images of high quality which pose a serious threat to existing Face Recognition Systems (FRS). In the literatures, it has been identified that FRS is vulnerable to multiform morphing attacks. Based on this vulnerability, several types of research on the detection of this morph attack was conducted using several techniques. Despite the remarkable levels of MAD reported in various literature, so far no suitable solution has been found to handle post-processed images such as images modified after morphing with sharpening operation that can dramatically reduce visible artifacts of morphed photos. In this work, an approach is proposed for MAD before image post-processing and after image post-processing built on a combination of Local Binary Pattern (LBP) for extraction of feature, Neighborhood Component Analysis (NCA) for selection of features and classification using K-Nearest Neighbor (KNN), Decision Tree Classifier (DTC) and Naïve Bayes (NB) classifier. The outcome gotten by training the different classifiers with feature vectors selected using the NCA algorithm improved the classification accuracy from 90% to 94%, consequently improving the general performance of the MAD.