Originally Posted by docrog
Could someone please explain the concept of "highlight recovery"? The term was recently used in the discussion of the S&M UHD disc when commenting on the horses in the snow. Thanks!
Originally Posted by DLCPhoto
I think of it in Photoshop terms, which while it might not be an exact analogy, probably comes close enough.
If you take a photo of a scene with a very wide dynamic range, the camera will have a hard time capturing the full range. If you're shooting Jpeg, the camera will process an image, and the whites may be clipped, with blown out highlights (think clouds for example). If you shoot in Raw Format, and take the image into Photoshop, you can use the Shadow & Highlights Tool to 'recover' the detail that is lost in the Jpeg, but still present in the Raw file, since it can capture a wider range than the Jpeg.
You can then make adjustments there, such that the highlights previously clipped are now recovered, with detail present, vs just detail-less whited out space.
I know nothing about how MadVR is doing its processing, but people speak of the 'Histogram' approach it uses as being somewhat revolutionary. Histograms are obviously of crucial importance when processing images in Photoshop, so this suggests to me that this analogy might not be too far off the mark.
- It sounds like you may be confusing Highlight Recovery with HSTM (Histogram-based DTM). Both these patent-pending algorithms help make madVR unique, but it completely different ways.
and all - The best way to describe Highlight Recovery is to consider the compression that happens to highlights as a result of tone mapping... Let's say you have an explosion or very bright highlights with two adjacent pixels, one encoded at 3000 nits and one at 4000 nits. This 1000 nit difference is very significant and would be clearly visible on a reference studio monitor that did not need tone mapping, as HDR masters may be typically encoded. Now, a tone mapping curve applies dramatically more compression at the top of the luminance range than it applies in shadow regions. Actually, it might not compress shadow regions at all. But it might apply a 1000:1 compression at the top of the luminance range. So 2 neighboring pixels with 3000 and 4000 nits after tone mapping might actually end up at 39 and 40 nits.
In that case there is a measly one nit difference in brightness between these two pixels, whereas the master had a significant different between the pixels. The result is that there is now no visible difference between these two pixels, and the same thing is happening across a huge number of pixels in the bright areas. This results in a significant loss of resolution in the brightest areas of the image, where we are most concerned about having great detail in a HDR image.
Our Highlight Recovery algorithm helps to prevent this crush, which results in much more defined and detailed highlight regions. In fact, it so effective that often times people incorrectly think that our Highlight Recovery is some sort of sharpening filter, but it is not - rather, it is just preventing the loss of detail that occurs in the highlights otherwise. So yes it does look a lot sharper and more defined, well, because it is, but not because we sharpened it, but because we prevented the loss in the first place.
This algorithm is very complex and to do it right and at our quality level requires a tremendous amount of GPU power that only madVR can deliver.
Likewise madVR uses other such techniques to restore damage to the luminance channel, to shadow detail, and contrast, but that's a bit off topic and too much to explain right now.