Painting creation method by impression feedback to a painting generator from an impression estimator
Tetsuro Sato, Kotaro Hayashi, Ikuo Mizuuchi · 2016
Is it possible that paintings created by an artificial agent touch a human? We investigate how an artificial agent “creates” paintings and what elements of paintings influence the human feeling. This paper presents the challenge of creating paintings by an artificial agent: a painting creating system. We propose the impression feedback so that an artificial agent can create impressive paintings, not based on retouching or other modification methods. The painting creating system draws a painting and predicts human impressions from the painting. Based on the human impressions, the system redraws another painting repeatedly. The method we propose consists of two modules: a painting generator and an impression estimator. The painting generator creates original paintings with brushstrokes. The impression estimator sees paintings and predicts human impressions with neural networks. This paper describes the method and the modules to construct the painting creating system.