It has become clearer where the similarity measure is picking up penalties that should be avoided for optimal performance. If the objective function was changed to apply weighting to the different points, then a lot of problems would be averted. For instance, earlier systematic experiments showed that there is more discriminative power around the nose, yet with two eyes equally accounted for, only a third (or less) of the overall similarity measure is based on outward geodesic dilation from the centre of the nose.
|
|
|
|
Roy Schestowitz 2012-01-08