|
Abstract : |
In this paper, we propose two adaptive interlaced-to-progressive conversion techniques in which the adequacy of the estimated motion vector is evaluated. If the motion vector is unlikely to give a good temporal motion compensated interpolation result, spatial interpolation is favored or selected to avoid temporal artifacts. In the first proposed interlaced-to-progressive conversion technique, called spatio-temporal weighted adaptive interlaced-to-progressive conversion, the interpolated value is a weighted sum of four interpolation filter results: the result by spatial vertical interpolation, the result by spatial directional interpolation using steerable filters, the result by temporal interpolation without motion compensation and the result by temporal motion-compensated interpolation. The most favored interpolation result of these four will receive the highest weight. In the second proposed technique, called similarity adaptive interlaced-to-progressive conversion, the spatial directional interpolation using steerable filters will be selected if it is likely to yield a better interpolated result than the temporal motion-compensated interpolation. The selection is done based on a similarity test. Our subjective viewing showed that these two interlaced-to-progressive conversion techniques correctly identify badly estimated motion vectors and occlusion areas. The use of spatial directional interpolation using steerable filters avoids the loss of image resolution encountered when shift-invariant spatial interpolation is used. 1., |
