FNRS2004 – “Swiss National Research Foundation Project: Real-Time 3D Hair Reconstruction”

Overview

Let us describe a scenario of the future that inspires our research program: A person is going to a ‘virtual’ hair dressing room. She is sitting on a chair while a number of pictures are taken from her head. Then she sees her virtual counterpart on a screen. Remarkably, the virtual counterpart is given with the hair captured from her, which takes physically plausible movement at an interactive rate. Now she selects other hairstyles from a 3D hair catalogue one after the other, which are automatically applied to her virtual counterpart. After she decides how she likes to have her virtual counterpart, she interactively adds further modifications. In a further more advanced scenario, she is able to view her real self on the screen, augmented with virtual hairstyles that are seamlessly integrated with herself, and react interactively to her movements as if they were her real hair.

MIRALab’s contribution

Our aim is to develop fundamental research that constitutes the realization of such ‘virtual hair-dressing room’ scenario. Although recent years have witnessed increased interest in hair modeling and simulation in the computer graphics community, the difficulties in simulation of hair, arising from the huge number of hair and its geometric intricacies, remains a challenge. The task demands even more effort when it is required to perform at an interactive rate, which means applying the style modifications on the model and computing its natural-looking movement and optical effects for the relative motion between hair and the light in real time. In the perspective of the scenario above, we propose the development of a new hair modeling and simulation technology that will lead a stride towards a previously impossible task. The framework is articulated into four separable problems. First, we discuss the hair modeling from images, which allows for rapid, automatic 3D reconstruction of the subject’s face and hair models. It distinguishes itself from existing hair model acquisition methods in that it aims at a unique, timely generation of hair model without having to rely on the user’s expertise and patience. Real-time simulation is one key technique that requires considerable development. Despite our years of expertise in physics-based simulation that can now be considered mature, the goal of real-time and interactivity sets very high constraints on the efficiency of the simulation system. The complex interaction of light and shadow among the hairs, and the small scale of thickness of one hair compared to the rendered image make the task of real-time hair rendering in MR (mixed realities) difficult but are essential for realism. Finally, real-time hair styling technique is essential to achieve hair with various styles almost instantly, which is a previously impossible task.

Partners

MIRALab, University of Geneva
Switzerland
www.miralab.ch