Abstract:

As a typical interactive color image segmentation method, the Grabcut algorithm is an important technique in the field of computer image. However, with the arrival of big data age, the types and quantities of image data are increasing exponentially. The workload of image segmentation is significantly increasing, and a higher demand is raised for the efficiency of image segmentation algorithms. Aiming at the problem of low efficiency and accuracy of the GrabCut algorithm, we propose an improved OneCut interactive image segmentation algorithm. Firstly, the energy function is built with the OneCut L1  distance term to avoid the Nphard problem faced by the GrabCut algorithm. Secondly, the apparent overlap penalty in the energy function is improved and the network structure is optimized by the color histogram acceleration technique. Finally, the complexity of the network diagram is reduced and the image segmentation efficiency and accuracy are improved. Experimental results show that the improved OneCut interactive image segmentation algorithm can significantly improve the segmentation efficiency and segmentation accuracy and a better segmentation result is obtained.

Key words: image segmentation, One-Cut, mincut, appearance overlap penalty term, GrabCut