Makeup Removal for Face Verification Based Upon Deep Learning

Makeup, derived from human`s pursuit of beauty, is widely accepted by the public. It changes the image of human`s appearance, brings more beautiful enjoyment and spiritual pleasure. Despite its popularization, it poses a huge challenge for face recognition, as altering the appearance reduces the accuracy of face recognition. In this paper, the purpose of the experiment is not only to generate non-makeup face images from makeup face images, but more importantly to retain identity information for face verification. To begin with, the original convolution layer is replaced by Resnet blocks. Furthermore, the idea and calculation method of feature matching are quoted. Experimental results demonstrate that this proposal generates non-makeup faces with few artifacts, that achieve 97.1% accuracy on Dataset1 and 94.3% accuracy on Dataset2 in face verification, which are better than discussed models.