Segmenting natural images with the least effort as humans

Great approaches to natural image segmentation have been made in recent years by learning from human segmentations, however little attention is paid to the behavior of human subjects in segmenting images. The paper investigates the effort made by human subjects and proposes an empirical method to estimate the boundary tracing loads, then establishes a model for natural image segmentation based on the least effort principle. We sort the hierarchies exhibited in human segmentation processes which use the BSDS tool, together with the monotonicity observed in the region merging processes, into two constraints on our model. Then an algorithm is established to segment natural images from scratch with pretty high efficiency thanks to the monotonic merging strategy. The experiment on BSDS500 shows our method obtains the state-of-the-art performance on both boundary and region measures. The average time consumption is only 1s and far less than those of its competitors. We also propose a new integrating evaluation measure, on which the performance of our method is noticeably worse than that of human subjects, indicating it is still a long run to build a perfect segmentation method.