Parametric fur from an image

Abstract

Parametric fur is a powerful tool for content creation in computer graphics. However, setting parameters to realize the desired result is difficult. To address this problem, we propose a method to automatically estimate appropriate parameters from an image. We formulate the process as an optimization problem wherein the system searches for parameters such that the appearance of the rendered parametric fur is as similar as possible to the appearance of the real fur. In each optimization step, we render an image using an off-the-shelf fur renderer and measure image similarity using a pre-trained deep convolutional neural network model. We demonstrate that the proposed method can estimate fur parameters appropriately for a wide range of fur types.