Importance of Data Term Normalisation
This page provides additional results that illustrate the importance of data
term normalisation in the proposed optic flow energy formulation.
As we will see, the results without normalisation either suffer from unpleasant
artefacts (if a small smoothness weight α is chosen), or cannot capture
the motion of the clouds (if a large smoothness weight is chosen).
To visualise the HDR images, we applied the tone mapping operator of
Fattal et al. as implemented
in the
pfstools package.
Exposure series Clouds
(512×340 pixels, exposure times between 1/250 and 1/30 seconds)
Download the exposure series as zip archive.
Result with normalisation: No artefacts are visible, and the motion of the clouds is nicely
captured.
Left: Corresponding tone mapped HDR result.
Right: Corresponding flow field between image 3 (reference) and image 4.
The flow is visualised by the colour code shown in the lower left corner.
Note that the images are given in full resolution, so you can zoom in!
Result without normalisation and using a small smoothness weight (α = 2.0):
Unpleasant artefacts occur at image edges.
Left: Corresponding tone mapped HDR result.
Right: Corresponding flow field between image 3 (reference) and image 4 of the series.
Note that the images are given in full resolution, so you can zoom in!
Result without normalisation and using a large smoothness weight (α = 10.0):
The motion of the clouds cannot be captured, resulting in unrealistic results in the top of the HDR result.
Left: Corresponding tone mapped HDR result.
Right: Corresponding flow field between image 3 (reference) and image 4 of the series.
Note that the images are given in full resolution, so you can zoom in!
<Additional Results for HDR Alignment
Main Page
Limitations>
|