Fast Bilateral Filtering for the Display of High-Dynamic-Range Images Frédo Durand & Julie Dorsey Laboratory for Computer Science Massachusetts Institute of Technology

Tone mapping 10-6 106 10-6 106 Pixel value 0 to 255 How can we display it? Linear scaling?, thresholding? 10-6 106 dynamic range Real world radiance 10-6 106 Display intensity Pixel value 0 to 255 CRT has 300:1 dynamic range

High Dynamic Range Images Here’s an HDR image that we constructed from the photographs you saw on the first slide. We’ll use this example throughout our talk. The dynamic range here is 25,000:1, which is much higher than the dynamic range of this video projector, so to visualize this image we show the logarithm of the luminance in pseudo-color. This image is very rich in detail, which is spread across the entire dynamic range. (click) For example, this slice of the dynamic range contains much detail in the darker regions of the scene, such as the leaves of the plant and the bulletin board. (click) This slice contains the details on the shadowed stone tiles outside (click) And this contains the luminances in some of the sunlit regions. Summary sentence: it can be a challenge to try to display all this wealth of visual information in a single image. low

Eye is not a photometer! "Every light is a shade, compared to the higher lights, till you come to the sun; and every shade is a light, compared to the deeper shades, till you come to the night." — John Ruskin, 1879

A typical photo Sun is overexposed Foreground is underexposed

Edge-preserving filtering Blur, but not across edges Anisotropic diffusion [Perona & Malik 90] Blurring as heat flow LCIS [Tumblin & Turk] Bilateral filtering [Tomasi & Manduci, 98] Input Gaussian blur Edge-preserving

Comparison with our approach We use only 2 scales Can be seen as illumination and reflectance Different edge-preserving filter from LCIS Large-scale Detail Output Compressed

Our approach Do not blur across edges Non-linear filtering Large-scale Output Detail Color