HDR in the Natural Scene

 

Figure 3.1 These six photographs were take over a period of one hour. The scene is constant; only the illumination changes.

The photographs in Figure 3.1 were taken at Bryce Canyon, Utah. They do a reasonable job of recording what we saw, sitting on a bench for an hour. A snow storm was approaching from the south. Bryce Canyon is at 9,000 feet elevation, so the clouds were close to the ground and moving very quickly. The materials in the image were quite constant (over eons).  The light changed rapidly and had both sharp edges from low clouds and gradients that were less visible. Observing the scene showed the variability of appearances from changes in illumination.







Figure 3.2 shows the photographs taken close to sunset and sunrise.


Figure 3.2 shows two photographs taken at the end of the day with increasing length of shadows (top row). The bottom pair of photographs were taken the next morning slightly after sunrise (left) and an hour later, with the arrival of the snowstorm. We can all recognize that the materials of the rocks are constant. The objects in the scene do not change. Any of the reproductive media described earlier can easily reproduce the dynamic range of the rocks. None of that media can accurately reproduce the combination of the material and the dramatic changes in illumination. The range of light in these scenes is greater than the range that cameras and humans can accurately record (Section B). Nevertheless, we can see details in the shadow, at the same time as seeing structure in the clouds. Just as our visual system can render HDR sensations of the world, we can synthesize low-dynamic range renditions of visual appearances. What we see is, in fact, a synthesis of the spatial content of the scene, rather than an accurate measurement of scene radiances. Successful reproductions render equivalent spatial information.

As we will see in Section B, it is not practical to capture and reproduce actual scene radiances.  We need to generate the equivalent appearance of the scene. To do that we need to understand how we generate appearances.