Qtpfsgui Notes The Durand TMO

This is the third post in this series - see also posts on the Drago and Manntiuk operators

Qtpfsgui Durand Notes

Test Image: Test1.hdr (Hass Burn images 357, 359, 361, April 2009 copyright DARW)

Initial Set-Up: As per Drago Notes, selecting Durand from the drop-down menu for TMO's.

Durand Tone Mapping Operator

According to the source http://osp.wikidot.com/parameters-for-photographers , this TMO is said to give the most realistic images with good detail.

The TMO menu gives three parameters:

Spatial Kernal Sigma, default 8.00, range 0 – 60.0,
Range Kernal Sigma default 0.40, range 0.01 – 10.00, and
Base Contrast, default 5.00, range 0 – 10.00.

There are, of course, the pre-gamma slider and the Adjust Levels tool to add to the options.

With Qtpfsgui 1.9.3, on the test image, the default settings give a totally blown white-out. Indeed, there appear to be no settings which give anything but a blown white image. Consequently, I am inclined to think that there is a fault in the implementation of the Durand TMO in Qtpfsgui 1.9.3.

Update: A very low pre-gamma setting (0.02 – 0.05) gives some image, almost all B/W with some yellow/green. Highly critical upon the exact pre-gamma value.

Reverting to Qtpfsgui 1.8.12, the Durand TMO works. (Note that the TMO panel is laid out in a different manner, but all the controls are present.) Using the default settings, the Durand TMO provides a slightly unsaturated result with blown highlights. Lowering the pre-gamma to 0.5 results in a more saturated image with some detail in the dark tones, but the highlights are still blown. Lowering the Base Contrast to 4.00, (default pre-gamma) gives an image with loss of saturation, but good histogram coverage, highlights are now good. Combining a lowering of pre-gamma to 0.5 and this lower Base Contrast gives a well saturated pleasing image with detail in the darks and detail in the highlights.

Keeping these settings, but changing Spatial Kernel Sigma from minimum (0.00) up to its maximum (60.00) does appear to do anything dramatic, perhaps a slight refinement of colour and sharpening at higher values. Increasing the Range Kernel Sigma, however, gives a more contrasty result reminiscent of OTT HDR images. Setting Range Kernel Sigma to 0 causes the program to crash and to values lower than its default artefacts start to appear.

Two further points about the Durand TMO should be noted. First, processing time for higher image sizes becomes substantial, 2048x3072 takes about 2 minutes on my system compared to about 10 seconds for 512x768. Second, this TMO exhibits another of the quirks of several Qtpfsgui tone mapping operators, namely size dependency. Larger file sizes often shown distortions such as blown highlights not apparent in lower file sizes.

Overall, setting Base Contrast at 0.4, pre-gamma at 0.5 and leaving the other two parameters at their default values results in a pleasing realistic image with good tonal coverage according to its histogram. Refinements can, of course, be made with Adjust Levels. OTT images can be generated by using extreme settings of Base Contrast and Range Kernel Sigma, but they do not appear to be particularly intriguing compared to, say, those from the Manntiuk TMO.



Technical Notes:

There are substantial papers on this technique by the authors F Durand and J Dorsey and S Paris available here http://people.csail.mit.edu/fredo/PU...dBilateral.pdf and here http://people.csail.mit.edu/sparis/p...roximation.pdf.

The method is said to be a fast bilateral filtering method for display of HDR images. In essence, if I understand the authors correctly, an HDR image is regarded as having a very wide range of different contrasts and the problem of display (as a low dynamic range image) depends then on mapping those differences in contrast. In order to preserve detail the authors split the HDR image into two layers, a base layer that encodes large scale variations in contrast, and a detail layer that encodes fine changes in contrast (i.e. detail!). The mapping process, which involves what the authors describe as a robust bilateral filer, is carried out on the base layer thus preserving detail in the final image. The parameters in the TMO, Spatial Kernel Sigma and Range Kernel Sigma, refer to variables that affect the strength of the application of the bilateral filter. Base Contrast refers to the degree to which the contrast of the base layer is changed.

Conclusion: Of no use in the most recent version of Qtpfsgui, but applicable with version 1.8.12 (and 1.8.8) giving good realistic images provided Base Contrast is set to a value lower than default and pre-gamma to about 0.5.

http://backup.cambridgeincolour.com/...044ded80_b.jpg



Cheers

David