Illumination Brush: Interactive Design of All-frequency Lighting Makoto Okabe ∗ Yasuyuki Matsushita † Li Shen † Takeo Igarashi ∗,‡ ∗ The University of Tokyo † Microsoft Research Asia ‡ SORST, JST Abstract We present an appearance-based user interface for artists to efficiently design customized image-based light- ing environments. 1 Our approach avoids typical iterations of parameter editing, rendering, and confirmation by pro- viding a set of intuitive user interfaces for directly specify- ing the desired appearance of the model in the scene. Then the system automatically creates the lighting environment by solving the inverse shading problem. To obtain a re- alistic image, all-frequency lighting is used with a spheri- cal radial basis function (SRBF) representation. Rendering is performed using precomputed radiance transfer (PRT) to achieve a responsive speed. User experiments demon- strated the effectiveness of the proposed system compared to a previous approach. 1. Introduction Image-based lighting, or the environment map, is a method of representing a large-scale lighting environment around a target scene as a texture map [5]. It compactly rep- resents complicated incoming light from distant sources and enables real-time rendering of the scene with realistic light- ing effects [28]. However, most systems rely on captured environments for image-based lighting [6, 10], and few at- tempts have been made to manually design such complex lighting environments. The artist might paint the environ- ment map directly, but this is labor-intensive and makes it very difficult to obtain the desired rendering result due to the non-intuitive relation between lighting conditions and the appearance of objects. Because a captured environment is not always available, a practical method for manually de- signing complicated lighting environments is in great de- mand. We propose an appearance-based user interface for de- signing image-based lighting environments. Instead of placing lights in the surrounding environment, users can directly specify the appearance of the resulting image by painting and dragging the color of outgoing radiance on the target model. Then the system constructs an appropriate image-based lighting model by solving the inverse lighting 1 This work was done while the first author was visiting Microsoft Re- search Asia. (a) (b) (c) (d) Figure 1. Designing image-based lighting with Illumination Brushes. The bunny model has a measured white glossy BRDF.(a) The user paints the desired diffuse appearance directly on the 3D model. Pink and orange diffuse brushes are shown. (b) The scene is rendered using the estimated environment map. The user paints a blue highlight on the bunny with specular brush. (c) All of the painted lighting effects are satisfied by ren- dering the bunny using the designed image- based lighting environment shown in (d). problem. Other appearance-based interfaces for lighting de- sign have been proposed [19, 24, 20, 12, 14, 1, 26, 2, 18], but all of them are limited to simple lighting models, such as point or directional lights, and are not designed for image- based lighting, which can produce more appealing results than simple lighting. It is particularly useful for adding real- istic lighting effects to synthetic objects when compositing them into a photographed or video-recorded background. Our system also supports the design of high-dynamic range (HDR) lighting on a standard low-dynamic range (LDR) monitor by introducing interactive tone-mapping, in which users specify a region of interest (ROI) and the system au- tomatically adjusts the tone-mapping parameter. Given the surface appearance specified by the user, the system must estimate the corresponding image-based light- ing model. This process is formalized as an inverse lighting problem, that is, recovering unknown lighting from known geometry, outgoing radiance, and the surface bidirectional reflectance distribution function (BRDF) [21]. To solve this problem at a responsive speed, we represent an image-based lighting model using the spherical radial basis functions (SRBF) [29], and estimate coefficients in a precomputed