I made a new operator for tone mapping, and I need to know it's efficiency (image quality) compared with other tone mapping operators.

  • 2
    What properties do you use to define image quality? Sharpest, histogram, etc? – Hueco Mar 2 '18 at 20:20

Due to the large difference between the original LDR and the resulting HDR image previously mean opinion scores were relied upon to evaluate the various algorithms, that proved very labor intensive.

In 2017 Aygül Salahl wrote an 87 page Master's thesis on his efforts to enlist the aid of two dozen subjects to evaluate four HDR tone mapped images in his paper "Comprehensive Study on High Dynamic Range Tone Mapping with Subjective Tests", this shows the work that goes into obtaining mean opinion scores.

To compare one HDR image to another the HDR visible differences predictor (HDR-VDP) algorithm has been used, see two versions here and here. What that accomplishes is to compare one altered image to another without regard to the original.

Previously it's been difficult to compare an HDR image to an LDR image due to the difference in dynamic range and some distortion imposed by the various algorithms; relying on PSNR or SSIM or other traditional metrics proved unreliable.

Efforts to alleviate this situation was undertaken by Jatollah Yeganeh and Zhou Wang of the University of Waterloo Canada, they came up with their "Tone Mapped image Quality Index" (TMQI) based upon a modification of SSIM and have released Matlab source code for their algorithm.

This was written about (.PDFs) here and here in Yeganeh and Wang's papers.

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