Automatic Object Segmentation Based on GrabCut

Abstract

Object segmentation is used in multiple image processing applications. It is generally difficult to perform the object segmentation fully automatically. Most object segmentation schemes are developed based on prior information, training process, existing annotation, special mechanical settings or the human visual system modeling. We proposed a fully automatic segmentation method not relying on any training/learning process, existing annotation, special settings or the human visual system. The automatic object segmentation is accomplished by an objective object weight detection and modified GrabCut segmentation. The segmentation approach we propose is developed only based on the inherent image features. It is independent with various datasets and could be applied to different scenarios. The segmentation result is illustrated by testing a large dataset.