The general field of this study is digital signal processing applied to imagesequences for multimedia applications. This work is divided into two maincontributions: an algorithm to segment images into moving video objects and atemporal interpolation method working with those objects.The segmentation of the sequence is performed with a temporal trackingalgorithm. A spatio-temporal segmentation algorithm is used to obtain initialregions in the first image of the sequence. This partition is then trackedwith an active contours technique, which operates on a novel segmentationrepresentation composed of open boundaries between regions. The algorithmestimates both the motion of boundaries and the motion of regions. It is alsoable to track multiple objects simultaneously and to handle occultationsbetween them. Results obtained on real image sequences show that thisalgorithm achieves a good temporal stability of the segmentation and a correctaccuracy of the boundaries.The goal of the interpolation algorithm is to reconstruct frames between twoimages in a sequence. It is a low-complexity algorithm which can be used atthe end of an object-based coder/decoder chain. The interpolation ismotion-compensated, and uses the motion of regions, estimated during thetracking. It is also object-based in the sense that the segmentation is usedto accurately predict occultation areas. This algorithm can be used in threedifferent applications: interpolative coding (where known images are predictedby interpolation), adaptation of the frame-rate to the terminal display in amulticast transmission and reconstruction of missing frames (where additionalframes are computed). Experimental results for the first application show thatfor a given reconstruction quality, the average compression is higher whenusing interpolation than with a causal prediction.
