Video Object Segmentation Using Spatio-Temporal Information and Marked-Watershed Operation

In this paper, we propose a video object segmentation algorithm based on spatio-temporal information and marked-watershed for extracting the moving objects from a video sequence. The algorithm begins with difference image between two adjacent frames and, using the Canny operator on the difference image, determines the initial edge mask for the object in motion. Morphological techniques are applied to the initial edge mask to obtain initial temporal segmentation mask of the moving object and binary marker image of the foreground and background subject to the watershed thresholding. The markers are used to modify multi-scale morphological gradient image of current frame. Finally, the watershed algorithm is performed on the modified gradients to locate the non-stationary objects accurately in the spatial domain of motion frames. Simulation results show that the proposed technique can overcome the shortcoming of over-segmentation of the watershed algorithm and can efficiently segment and extract meaningful motion objects with slow or fast moving from the video sequence with complex background with low computational complexity.