Portraits Analysis

June 5, 2019 , Israel. The full paper by Emily Drake (Siggraph)

Researchers at the Interdisciplinary Center in Israel have developed a technique to automatically analyze portrait artwork, factoring in fine-grained detail, accuracy, and individual artistic style. The research aimed to furnish a computational methodology for detecting facial features in artwork, or facial landmarks like eye or mouth corners. The Efi Arazi School of Computer Science's Ariel Shamir said, "Our key idea was to create such data using what we call 'artistic augmentation.' We transform photographic face data to be more similar to artistic portraits and use it to train new models for neural networks that work better for artistic inputs." The researchers assessed their detection technique by producing a dataset of artistic faces, featuring 160 portraits by 16 artists in various genres and styles, with wide variance in geometry and texture.