【摘要】 目前各种显著性区域检测算法使用的训练数据大多源于一般性视觉注意过程，主要反映的是对图片中明暗变化信息的注意，没有专门针对图片的颜色信息获取训练数据，因此训练出的算法主要反映明暗信息变化以及形状等方面的显著性。然而，颜色为视觉提供了更多有用的信息，图片显著性检测算法应该也必须考虑颜色因素。针对此问题，设计颜色注意实验，通过实验获取训练数据，用于对PiCANet、PoolNet、U2-Net以及BASNet显著性检测深度学习模型进行训练与分析。结果显示，颜色信息的加入能够使算法获得更好的效果。对比不同算法，BASNet能够较好地检测图像颜色显著性区域。更多还原

【Abstract】 At present, most of the training data used by various saliency region detection algorithms are derived from the general visual attention process, which mainly reflects the attention to the light and dark change information in the picture. There is no training data specifically for the color information of pictures. Therefore, the trained algorithm mainly reflects the saliency of the light and dark information change and shape. However, color provides more useful information for vision, and image saliency detection algorithms should and must take color into account. In response to this problem, this paper designed a color attention experiment and obtained training data through the experiment for training and analysis of PiCANet, PoolNet, U2-Net and BASNet significance detection deep learning models. The results show that the addition of color information can achieve better results. Compared with different algorithms,BASNet can best detect the color saliency region of the image.更多还原