摘要
针对现有图像显著性区域提取算法都是以图像像素为基本单元进行计算,因而会导致显著性表示中一致性较差、运算量较大等问题,提出一种新的图像显著性建模方法,即以超级像素为基本单元,提取颜色直方图、区域纹理等特征,融合整体比较模型和局部比较模型,有效地解决显著性表示一致性的问题.实验表明,通过对标准图像数据集的显著性提取、分割,该算法能够有效地表示颜色、密度等方面的显著性;与现有算法相比,获得的图像显著性表示具有更好的一致性效果.
Most algorithms compute saliency based on local features and they usually fail to uniformly highlight a whole salient region or handle complex computation. A novel model was proposed to compute saliency maps based on region features, and a globak framework was constructed to combine region features, such as color histogram and texture. Compared with the models based on local features, the results of the experiments on image datasets demonstrate that the proposed model performs much better in describing visual saliency caused by color and texture and is more efficient in uniformly highlighting salient regions.
基金
国家自然科学基金(61003136)资助