Synthesizing Face Images from Match Scores

In this paper, we consider the problem of generating a face image based on its match scores with other face images. Such an exercise is not only useful in understanding the relationship between face images, but it can also be used to understand the degree of privacy associated with match scores. We address the problem using two approaches. The first mixes face images to deduce the appearance of a missing face image and the second uses a convolutional auto-encoder to further enhance the mixed face image. Experiments suggest the potential of the proposed approaches in generating a missing face image in a database by utilizing its relationship with other images in the database.