
Below are the simulation results for the Brightside encoder (BS) and the convex optimization encoder (CO), compared to the 'ideal' result generated by rescaling the source HDR image to fit the Brightside panel. All images were normalized to have a mean of 1, and then a sequence of exposures was generated to allow viewing of the entire dynamic range. The exposure factors are powers of two, with the number in the boxes below indicating the exponent. The final two images are the mean-square error images comparing the encoded images to the source. These are log-scaled to allow the full dynamic range to be visible, with the error of the color channels for each pixel averaged together.
| CO | Ideal | BS | |
|---|---|---|---|
| -8 | -8 | -8 |
| -6 | -6 | -6 | |
| -4 | -4 | -4 | |
| -2 | -2 | -2 | |
| 0 | 0 | 0 | |
| 2 | 2 | 2 | |
| 4 | 4 | 4 | |
| Err | Err |
The error between the ideal and encoded images here becomes very large at both the skylight and the rightmost stained-glass window. These are the regions with the highest luminance, and it seems that it exceeds the dynamic range the encoders can successfully code for. Therefore, the error grows very large since the HDR display saturates at a brightness signficantly below what would be required to reproduce the actual dynamic range. This gives rise to a high average error (15.5 for CO, 46.79 for BS), but the median error is still very low (0.0021 for CO, 0.0054 for BS). With the final code, the CO encoder has both better mean and median error than the Brightside encoder,
It took the BS encoder 11.1 seconds to process this scene, and the CO encoder 277.9 seconds (A slowdown of a factor of 25)
| CO | Ideal | BS | |
|---|---|---|---|
| -8 | -8 | -8 |
| -6 | -6 | -6 | |
| -4 | -4 | -4 | |
| -2 | -2 | -2 | |
| 0 | 0 | 0 | |
| 2 | 2 | 2 | |
| 4 | 4 | 4 | |
| Err | Err |
This scene has lower overall dynamic range than the Memorial Church scene, and the errors are correspondingly far less. The encoded images can faithfully represent the dynamic range in all areas of the image.
| CO | Ideal | BS | |
|---|---|---|---|
| -8 | -8 | -8 |
| -6 | -6 | -6 | |
| -4 | -4 | -4 | |
| -2 | -2 | -2 | |
| 0 | 0 | 0 | |
| 2 | 2 | 2 | |
| 4 | 4 | 4 | |
| Err | Err |
This scene once again has a very high dynamic range, so the encoded images cannot quite capture the full range. Again, this leads to errors at the high end, which are unlikely to be perceptually significant.