分割评价是改善算法性能的重要途径。针对当前图像分割评价指标不能很好地反映分割结果的问题,提出组合赋权的灰色评估模型。首先,在现有典型评价准则中选取概率边缘指数、全局一致性误差、变换信息量3个准则来评价图像分割质量。其次,提出结合德尔菲法、强制判定法和熵权法的主客观组合赋权法,使权重既反映观察者主观偏好,又突显图像客观差异。最后,利用所提出的模型对测试图像进行综合评价。实验结果表明,所提出的评价模型更符合主观评价结果与地面真实结果。将此模型用于比较基于花粉算法、遗传算法、蛙跳算法的最大熵阈值算法所得到的分割图,与最大熵的排序结果一致,进一步验证了该模型的有效性。

Segmentation evaluation is an important way to improve the performance of algorithms. A gray evaluation model is proposed based on combinational weighting, aiming at the problem that the current index of image segmentation can not reflect the results of segmentation well. Firstly, variation of information, global consistency error, and probabilistic rand index are selected to evaluate the quality of image segmentation. Secondly, a subjective and objective combinational weighting method is proposed which combines Delphi method, forced decision method, and entropy method. The weights not only reflect the subjective preferences of observers, but also highlight the objective differences of images. Finally, the proposed model is used to make a comprehensive evaluation of test images. Experimental results show that the proposed evaluation model is consistent with the subjective evaluation results and the real ground results. Moreover, this model is used to compare the segmentation results of the maximum entropy threshold algorithms based on flower pollination algorithm, genetic algorithm, and shuffled frog leaping algorithm, respectively. The obtained rank is consistent with the result of maximum entropy, which further validates the effectiveness of this model.