A moving object identification algorithm for image sequence interpolation

An effective motion compensated image interpolation technique is presented. The proposed algorithm has been developed for the interpolation of missing frames in an image sequence and it is based on two principal elements. The first is a region-based motion estimation and representation technique, which combines, for each known image, an approximate initial motion field estimate and an initial image over-segmentation (obtained using both luminance and chrominance information) to produce a very accurate affine-model regularized motion field. The proposed algorithm uses a robust identification technique for the estimation of the motion parameters associated to each region. The second important element is the use of an interpolation technique, that starting from the available motion information defines, for each point of the image to be interpolated a suitable reconstruction strategy taking into account the different situations that can appear (for example a moving object can occlude either a stationary background or an another object and so on). The principal aim of the proposed algorithm is the reconstruction of the missing frames in a reasonable way, without introducing significant artifacts and assuring a pleasant representation of object displacement