基于Object Proposals并集的显著性检测模型

Saliency detection model based on the union of Object Proposals

针对当前常见的显著性检测模型得到的结果会包含大量的背景区域的缺点,本文提出了基于Object Proposals并集的显著性检测模型。该模型首先对于输入图片生成一系列Object Proposals,并通过其并集计算得到背景图;然后结合纹理特征和全局对比度得到初始显著图;最后,用得到的背景图对初始显著图进行背景抑制得到最终显著图。实验结果表明,在通用MSRA1000数据集上,本文提出的显著性模型与其他5种方法相比取得了很好的效果。

In saliency detection,current existing models usually produce results containing many background regions.To improve the performance,a novel saliency detection model is proposed based on the union of object proposals.The model first generates a series of object proposals from the input pictures,and then gets the background map by computing the union,and then obtains the initial saliency map by combining the texture and global contrast.Finally,the final saliency map is derived by restraining the initial saliency map with the obtained background map.Experimental results on the general MSRA1000 dataset demonstrate that the proposed saliency model performs well compared to the other five existing methods.