MDL-Based Coherent Motion Segmentation for Semantic Object Identification

The extraction of semantically meaningful objects that can describe a video sequence provides great expectations for video compression and retrieval. In this paper we propose an effective region merging technique that coherently merge moving image regions in a Minimum Description Length framework. This technique uses a reliability measure that indicates to what extent an affine parameter set represent the motion of an image region. To overcome the region-based motion estimation and segmentation chicken and egg problem, the motion field estimation and the segmentation task are treated separately. A preliminary motion field estimation is carried out starting from a traslational motion model. Concurrently, a Markov Random Field model based algorithm provides for an initial static image partition. For each image region of this spatial segmentation, affine motion parameters are then extracted from the motion field by means of a robust linear regression algorithm. A merging stage based on the proposed technique gives the final spatio-temporal segmentation.