Generative Portrait Sketching

We consider the problem of generating art with a computer system, as it relates to the understanding of a drawing style. Our system in its present development is able to sketch faces automatically, starting from a picture, typically a photographic snapshot of a scene with humans. Once a digital image is considered, the system automatically finds where some of the faces (or face-like patterns) are and isolate these. Each face image pattern is then used to produce a stylised portrait. The style which the system aims at using is derived from the one the first author has developed over the years in his artistic practice. The goal is for the final system to mimic the process developed by the artist, rather than aiming at results exactly reproducing his way of drawing a portrait. Nevertheless, the produced sketches are in the style of the artist. Various steps in the process of producing a sketch by the artist were isolated, including: image segmentation, filtering, shape selection and depiction, filling and shading. The implementation of these steps relies on an understanding of human visual perception, of the artist’s work process, and of advances made in computer vision.