Difference between revisions of "HDR workflow with hugin"

From PanoTools.org Wiki
Jump to: navigation, search
(Rewrote from scratch)
m (typos)
Line 17: Line 17:
[[Contrast Blending]] approach, but has the potential for misalignments causing
[[Contrast Blending]] approach, but has the potential for misalignments causing
[[ghosting]] problems.  The second technique is presented here since it
[[ghosting]] problems.  The second technique is presented here since it
involves a greater range of techniques.
involves a greater range of tools.
== Preparing the HDR images ==
== Preparing the HDR images ==
Line 163: Line 163:
=== Tone mapping ===
=== Tone mapping ===
[[tone mapping]] aperations use [[HDR compression]] to compress high [[dynamic
[[tone mapping]] operations use [[HDR compression]] to compress high [[dynamic range]] images.
range]] images.
A related package to [[pfstools]] called [[pfstmo]] can do automatic [[tone mapping]]
A related package to [[pfstools]] called [[pfstmo]] can do automatic [[tone mapping]]

Revision as of 15:50, 1 May 2006

This tutorial doesn't cover reasons why you might want to stitch in HDR format. It is a simple HOWTO listing the tools available and how to use them with hugin.

Working with HDR images is fairly extreme behaviour. If you just want higher quality output than you get with typical 8bit photography, then you probably want to look at a 16bit workflow with hugin.

Still here? There are two basic ways of creating an HDR panorama:

  • Stitch several panoramas of the same scene, each one at a different exposure, and merge them together into a single HDR file.
  • Create a set of HDR shots of the scene and then stitch them together.

Each has advantages and disadvantages: The first technique is simpler and has the advantage that the final HDR step can be skipped and substituted with a Contrast Blending approach, but has the potential for misalignments causing ghosting problems. The second technique is presented here since it involves a greater range of tools.

Preparing the HDR images

Unless you have an expensive HDR camera, you will be merging bracketed shots to create the HDR images - Unfortunately this means that you are limited to static scenes and landscapes.

Taking bracketed shots

The number of shots required depends on the dynamic range of the scene you need to capture and the capabilities of your camera.

Many cameras have an auto-bracketing mode that takes three or five shots two stops apart with one press of the button. This may be adequate, though a typical outdoor scene might have a dynamic range of eighteen stops which would require eight shots two stops apart.

Whatever method you choose, it should be obvious that you need a good tripod to keep the camera steady.

Merging bracketed shots with pfscalibration

There are other tools for merging bracketed images, but pfscalibration is Free Software and does the job.

Calibrating the camera response curve

Generally when a digital camera creates a JPEG or TIFF file, it takes a 12bit per-channel dynamic range image captured by the CCD and compresses it using a camera response curve into a 8bit output file.

So JPEG and TIFF files need unrolling with a calibrated camera response curve so they can be mapped into the linear space of the floating-point HDR image.

If you are working with RAW images, the camera response is generally linear and doesn't need calibrating, so you can skip this step.

A quick way to derive the response curve for later use is to take a series of five bracketed JPEG shots one stop apart. eg. 2, 1, 0.5, 0.25 & 0.125 seconds exposure. First extract the exposure times from the EXIF data:

 jpeg2hdrgen *.jpg > mycamera.hdrgen

Then extract the response curve, by comparing the photos, and save it:

 pfsinhdrgen mycamera.hdrgen | pfshdrcalibrate -v -s mycamera.response

By default this mycamera.response file contains a weighting table that effectively throws away the brightest and darkest pixels. So open the file in a text editor, find the weighting table at the end, and change the zeroed values (0.000000000) to a positive number (0.001000000).

Aligning the shots

Even with a tripod, unless you have a programmable camera, remote control or very steady hands; it is unlikely that your series of bracketed pictures align perfectly, so:

  • Start up a new hugin project for each series and load the images.
  • Create a few hundred control-points between each pair of consecutive photos with the g key in the control point tab.
  • Fine tune all points and delete any with a correlation less than 90% (ie. enter -0.9 in select by distance and delete).
  • Optimise positions and delete any control points with an error greater than 0.2 pixels, reoptimise.
  • Adjust the field-of view in the stitching tab to something slightly smaller than the input size so there are no transparent edges.
  • Stitch to multiple TIFF format.
  • Rename the output TIFF files to match the input images.

These TIFF files need to have the alpha channel stripped away, you can do this be converting to and from PPM format with ImageMagick:

 convert DSCN4804.tif DSCN4804.ppm
 convert DSCN4804.ppm DSCN4804.tif
 rm DSCN4804.ppm

Merging the bracketed images to OpenEXR HDR format

Create a hdrgen file listing each of your bracketed photos and their exposure times, you can base this on the mycamera.hdrgen file created earlier.

Then use this and your camera response file to create an EXR file:

 pfsinhdrgen mypicture.hdrgen | pfshdrcalibrate -v -f mycamera.response | pfsoutexr mypicture.exr

Check the output with pfsview:

 pfsinexr mypicture.exr | pfsview

Converting to floating-point TIFF format

hugin doesn't yet read EXR files, so you need to convert to TIFF.

pfstools should be able to create HDR 32bit floating-point TIFF files directly, but for some reason, this doesn't work for me. So open each EXR file in cinepaint and save as TIFF.

Correcting chromatic aberration

Now is a good time to correct chromatic aberration and vignetting with fulla using pre-calibrated data for your lens.

Stitching with hugin

The TIFF images can be loaded into hugin as per usual with a couple of caveats:

  • Everything will appear very dark since our images represent linear sensor data.
  • Information about the field of view was lost, so this will need to be re-entered manually or re-optimised with PTOptimizer.

Stitch the images as per usual into a TIFF file, you can use enblend as the final step.

Post processing

This TIFF file is in floating-point 32bit per channel IEEE format. This is impossible to display in its entirety at once, so you possibly want to create final 8bit per channel human-readable images.

Otherwise, typically a HDR panoramic image is used as a lightprobe for 3d rendering, in which case you are now done.

Adjusting in a GUI tool

There are various image editors that can open this file such as cinepaint, krita, vips and HDRIE. The capabilities vary so you need to experiment.

Alternatively, use pfstools to manipulate the image: pfstools has the facility to read HDR TIFF files, unfortunately it chokes on the alpha channel created by nona and enblend - Use cinepaint to open the TIFF file and save as EXR.

This can be viewed with the pfsview GUI tool:

 pfsinexr stitch.exr | pfsview

A quick way to create a usable 8bit per channel image is to select a gamma of 2.2, fit to dynamic range, zoom 1:1 and save as PNG.

Tone mapping

tone mapping operations use HDR compression to compress high dynamic range images.

A related package to pfstools called pfstmo can do automatic tone mapping of an HDR image and compress it into a low dynamic range output. There are many options and techniques available, commands look like this:

 pfsinppm stitch.exr | pfstmo_drago03 | pfsgamma -g 2.2 | pfsout stitch.png