Adventures in HDRI

Contrast is the bane of the photographer. The human visual system is highly adaptive in its response to light levels and hence our experience is a highly compressed version of the “real” scene. Image sensors have difficulty in recording large extremes in light intensity which may be present in a typical scene. For instance, sunlit cloud may be tens of thousands of times brighter than deep shadow so exposing for one will leave the other either completely black or burnt out. To give the same visual impression, we need to compress the variation of light level in a scene to a more manageable level.

ThereĀ are two steps to this, a) the capture of the image data, b) the display of the captured data. The first step is called high dynamic range imaging (HDRI) capture. Various options exist for the second step, though what most people call HDR photography is one option called tone-mapping.

HDRI capture usually involves taking am exposure bracket sequence of images to capture both the shadows and highlights of an image. Depending on the scene, this may require exposures -5EV to +5EV as well as ones in between to cover the gaps. These brackets are then merged into a single HDR file containing a scene-referred set of pixel values. To cope with the large range of possible values, either high bit-depth or floating point numbers are used.

Sunset over Pacific QuayAfter the creating of a HDR file, it is usual to create an output-referred file for display. Tone-mapping is a popular option by which many gaudy, surrealistic photos are created (I include an example to the left). This is what many people associate with the term HDR.

It does not have to be this way, many tone-mapping methods produce naturalistic looking images which are not obviously distinguishable from a “normal” photo (apart from being able to represent scenes of extreme contrast). Exposure fusing (my preferred alternative method) bypasses the production of a scene-referred HDR file and tends to give more natural looking images.

From my experiences in HDRI, the most important aspect is the capture and conversion of the bracketed images ready for merger. My tips are:

  • Exploit the natural dynamic range of modern digital sensors to bracket at 3EV. Smaller bracket intervals (e.g. 1EV) may be useful in reducing noise but are not strictly necessary with careful post-processing.
  • White balance during RAW conversion is extremely important. Incorrect white balance will lead to colour casts in the merged HDR file which can be inconsistentm especially detrimental to panorama creation.
  • During RAW conversion, try to linearise the output as much as possible, e.g. using minimum contrast and a “linear” curve. Most RAW converters (e.g. Lightroom) produce output files with gamma curves applied along with exposure tweaks. Choosing minimal contrast will increase the amount of image values falling in the middle of the exposure curve hence reducing the effect of the tone curve at the ends.
  • Try mid-tone emphasis (in Picturenaut) when merging files, especially when using minimum contrast conversion as above. This also reduces the effect of end effects of the RAW converter tone curve.