Capture and Processing - 3DTV: Processing and Transmission of 3D Video Signals - page 58

need blending. 3D images look bad if the two source images have different
colours. To overcome all such negative effects, the multi-view videos need
colour correction before depth estimation and coding.
A histogram-based pre-filtering method had been described in [16] that
is applied on the rectified multi-view camera images for colour correction.
This method comprises the derivation of a mapping function that adapts the
cumulative histogram of a distorted sequence to the cumulative histogram of
a reference sequence. In the case of multiple cameras, all camera views should
be adapted to a common reference camera view. The adaptation should be
done for all R, G and B channels separately for every frame. Figure 2.14 shows
a plot of the luminance histograms of the three sample images: the reference
image used to correct the colour of the distorted image, the distorted image
before colour correction, and the distorted image after colour correction. The
reference image comes from the common camera view, whereas all other
images in other camera views are regarded as distorted. Note that the bins
in the histogram of the distorted image have been modified to match the
distribution of the bins in the histogram of the reference image after colour
correction. The colour differences are more significant especially at the peaks
of the histogram, i.e. where the peaks of the reference and the distorted image
do not align. From Figure 2.13 it is clear that most such peaks in the distorted
Figure 2.13 Luminance histograms of the reference image, the distorted
image before and after colour correction