New rule-based framework for post-processing merging in video sequence segmentation

This paper addresses the problem of semantic object segmentation in a video sequence for multimedia applications. It presents a new framework to generate a segmentation from multiple segmentations of an image (e.g. using intensity, motion, motion projections from past frames, etc.). We define a rule processor that applies a set of merging and splitting conditions to the segmentations. Each condition is model by a rule which is characterized by two functions, one to verify compliance, and the other to apply the rule. This framework is flexible since rules can be easily added or removed from the processor. Results obtained with two sets of simple rules show the possibilities offered by this framework