Face texture synthesis from multiple images via sparse and dense correspondence

We have a desire to edit images for various purposes such as art, entertainment, and film production so texture synthesis methods have been proposed. Especially, PatchMatch algorithm [Barnes et al. 2009] enabled us to easily use many image editing tools. However, these tools are applied to one image. If we can automatically synthesize from various examples, we can create new and higher quality images. Visio-lization [Mohammed et al. 2009] generated average face by synthesis of face image database. However, the synthesis was applied block-wise so there were artifacts on the result and free form features of source images such as wrinkles could not be preserved. We proposed a new synthesis method for multiple images. We applied sparse and dense nearest neighbor search so that we can preserve both input and source database image features. Our method allows us to create a novel image from a number of examples.