Objectness based unsupervised object segmentation quality evaluation
Ran Shi, King Ngi Ngan, Songnan Li · 2017
In this paper, we propose an unsupervised objective measure for quality evaluation of single object segmentation in images. Objectness as an essential attribute of objects is treated as a main feature to measure object segmentation quality. In addition, the prior information about the object quantity is integrated into the proposed measure. Experimental results show that our measure can conform well to the subjective rating of the object segmentation quality.