Capture and Processing - 3DTV: Processing and Transmission of 3D Video Signals - page 61

to right to produce the disparity estimates in each camera. Both estimates
are accordingly compared against each other for a consistency check and to
produce a confidence kernel making use of the normalized cross-correlation.
In the successive adaptive tri-lateral median filtering stage, an adap-
tive filter is applied on the initially estimated disparity values for which
the weighting coefficients are determined based on three different factors:
the confidence kernel, the results from the colour texture analysis (i.e. the
similarity in luminance intensity of the centre pixel and the weighted pixel)
and the proximity (the distance from the centre pixel). Accordingly, at each
current pixel position s , where the filter is centred, a disparity intensity value
Figure 2.14 Disparity estimation results using the mentioned HRM method
(in the middle) and the successive ACTMF stage (at the bottom) for the left
camera of ''Caf e'' MPEG test sequence by Gwangju Institute of Science and
Technology (Reproduced with permission of Professor Yo-Sung Ho)