Watershed From Propagated Markers Improved by the Combination of Spatio-Temporal Gradient and Binding of Markers Heuristics

Abstract

This paper presents the improvement of the watershed from propagated markers, a generic method to interactive segmentation of objects in image sequences, by the inclusion of a temporal gradient to the segmentation framework. Segmentation is done by applying the watershed from markers to a gradient image extracted from the temporal gradient sequence and using markers provided by the binding of markers heuristics. The performance of the improved method is demonstrated by application of a benchmark that supports a quantitative evaluation of assisted segmentation of objects in image sequences. Experimental results provided by the combination of temporal gradient with the binding of markers heuristics show that the proposed improvement can decrease the number of human interferences and the time required to process the sequences.