Demo paper learning to beautify facial image

摘要

In this demo, we demonstrate a data-driven facial image beautification system that learns how to beautify portraits from facial image database, and enhances the facial texture of arbitrary portraits automatically by modifying its pigment distribution and correcting its color. Specifically, as human skin can be treated as a turbid medium with multilayered structure, we decompose facial image into melanin and hemoglobin layers. With the extracted attractiveness features, a data-driven qualitative beatify model serves for the guidance of beautification through optimizing hemoglobin and melanin layers. Our beautification operations are conducted in completely automatic and time-efficient way, leading to customized realistic beautified portraits that follows users' preferences.