Binary conversion method based on illumination distribution model under unconstrained smartphone face database

One of the most important methods to extract unique features for distinguished several classes in face recognition is by generating a binary image from the original image, which is then used as input in the feature extraction process. In face recognition scenario, illumination variation is a challenging problem due to the dramatically changes of face appearance depending on the illumination conditions. It has a major effect on the clarity of binary image, where most of the binary image conversion methods had failed under the effect of these conditions. Therefore, to deal with this problem effectively, a model of illumination distribution over a whole image is proposed, this model is based on the polynomial function and it is used as an input for proposed binary conversion method which starts by calculating an adaptive threshold value depending on the illumination model and converts the image by comparing each column pixels individually with the adaptive threshold value. The proposed method works by making evaluation on smartphone face video dataset and comparing with the Global image threshold using Otsu's method. The experimental results showed the outperformance of proposed method.