基于信息量增强的共生直方图显著性检测算法

针对目前基于共生直方图显著性方法检测得到的显著性区域容易受到背景区域高对比度边界的干扰，提出了一种通过信息量增强的改进后的共生直方图检测方法。该方法利用显著性区域和背景边界区域在局部共生直方图分布复杂度差异较大的特点，并使用信息量值来量化这种差异。在进行显著度计算时将对局部共生直方图复杂度估计的信息量值叠加到原始算法的各个通道中，从而达到增强显著性区域的同时抑制背景边缘显著性的目的。对AIM数据集进行数值实验，该方法对原方法在准确性和鲁棒性上都具有较为明显的改善，通过ROC曲线进行分析，所提出方法将原方法的AUC值从0.720 8提高到0.731 1。

Co-occurrence histogram based saliency detection method enhanced by entropy

This paper proposes an image enhanced saliency detection method which overwhelms the shortage at the existing co-occurrence histogram based methods that are easily to be influenced by the high contrast edge object in background regions. The motivation of the method is inspired from the fact that the difference between the co-occurrence histogram distributions indexed from salient regions and background edge regions is very large. According to it, the method uses the entropy to describe the distribution complexity and measures their difference. In order to achieve the purpose of enhancing the salient region and inhibiting the edge of background region, the entropy is multiplied to the saliency value in original algorithm in each channel. The experiments on the AIM data set show that the proposed saliency model is more accurate and robust than original models. And the proposed model can improve the AUC value from 0.7208 to 0.7311 through the ROC curve.