A Co-segmentation Method for Image Pairs based on Maximum Common Subgraph and GrabCut

The outputs of a super-pixel-level co-segmentation method for image pairs based on maximum common subgraph (MCS) often contain some background regions and can not completely cover the common objects. Based on this, for each image, we first discard redundant regions which are usually small. Then, we define pixels inside the minimum enclosing rectangle of remaining object regions as possible foreground. And the rest of pixels of the image are defined as background. Finally, we use these preliminarily classified pixels to initialize GrabCut algorithm and execute the algorithm to obtain the final results. Thus, a new pixel-level co-segmentation method which combines maximum common subgraph and GrabCut algorithm is developed like this. Our method is tested on a commonly used image pair dataset. And the results show that the proposed method can efficiently co-segment common objects from image pairs with high Precision 87.8% and Recall 87.7%.