I'm trying to learn how to merge several images to fix blown-out backgrounds, but keep things as natural as possible. The HDR merge feature seems very unnatural vs the blend layers looks very reasonable. Are there reasons the latter isn't used more frequently?

Original two test images:

enter image description here

enter image description here

HDR Imagelayer blend

HDR Layer blend

1 Answer 1


HDR is in its simplest form blending where the pixels are weighted by their distance from the middle value. Blending is more like a constant weight or other function based on blending mode. In your example you are not really gaining much extra DR in the outside area. Your HDR does recover teh details but have a flat look because of the boring tonemapping that needs to be done to map the floating point image back to 8bit. It looks like my technical images where I just apply a gamma curve and nothing else.

Remember your original 2 exposures already used your dynamic range in 8 bits, and now you squeeze these together. Make sure you get the result back in 16bits so you can control the tone mapping. Adjust gamma, curves, and saturation. you especially want to split process the saturation so teh darks get desaturated more than the bright areas.

HEre I did a bit of tweaking on your 8 bit HDR image:


For kicks I copied your two exposures into image view plus more to see how my own algorithm looks. I dont remember if I activated a color tone mapping based on human perception experiments I found in a research paper in the code at this moment, though. Id think yes, as the colours out of the box resemble more the colours if my tweaked version of yours, except for the dark areas and the green.

my algorithm

  • Thanks! When do you think one method is more appropriate than the other? Is blending more "natural" but HDR more blended and saturated? Jun 27, 2013 at 6:24
  • The basic HDR algorithm is the natural one as the formula is based on physics. The blending method is more of a hack that might be twaeked. The HDR only looks unnatural because there is a fashion for those oversaturated crazy results. It is not inherent in the method itself. My first algorithm was actually loosing the colour saturation a lot, which is hwy I searched for research in improving the color rendition in tonemapping, as boosting global saturation (which I think if the normal way) is not good. The human perception way was mapping the saturation boost to a sigmoid function Jun 27, 2013 at 8:39
  • and the found some constants through some tests on people that would evaluate how identical colours were with different models, and plotting the perceived quality and doing regression to the good parameters. Jun 27, 2013 at 8:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.