提出了基于运动纹理差异的运动目标分割无监督评价方法。通过基于贝叶斯决策的精确量化准则得到差分图像的类别映射图, 利用区域生长方法产生不同效果的运动目标分割掩膜, 定义了运动纹理差异的评价准则, 并将该评价准则用于这些分割掩膜中以验证评价结果的有效性。为进一步准确衡量所提出准则的性能, 在实验中将其与两种评价准则进行对比, 一是有监督评价准则 ME, 二是基于彩色纹理的无监督评价方法 JSEG中的无监督评价准则 J。实验结果证明基于运动纹理差异的无监督评价与人类主观感知和客观实际基本一致, 可有效地对红外视频运动目标分割性能进行无监督评价。

An unsupervised evaluation method for moving object segmentation based on motion texture difference was proposed in this paper. A class map of the difference image was obtained using the exact quantization criterion based on the Bayesian decision. Moving object segmentation masks with different effects were obtained using the region growing method. An evaluation criterion for identifying the difference between the motion textures was defined. Moreover, evaluation criteria were used to verify the validity of the evaluation results. The results were compared with two other evaluation criteria to accurately measure the performance of the proposed criterion. The two other criteria were as follows: the supervised evaluation criterion, ME; and the unsupervised evaluation criterion based on the unsupervised color texture evaluation method, JSEG. The experimental results showed that unsupervised evaluation based on the difference between motion textures was in agreement with human subjective perception and objective reality; this could be used to evaluate the performance of moving object segmentation for infrared video effectively.