Saliency-Detection

代码说明：

  提出一种新的显着性检测方法，通过将区域级显着性估计和像素级显着性预测与CNN（表示为CRPSD）相结合。对于像素级显着性预测，通过修改VGGNet体系结构来执行完全卷积神经网络（称为像素级CNN）以执行多尺度特征学习，基于该学习进行图像到图像预测以完成像素级显着性检测。对于区域级显着性估计，首先设计基于自适应超像素的区域生成技术以将图像分割成区域，基于该区域通过使用CNN模型（称为区域级CNN）来估计区域级显着性。通过使用另一CNN（称为融合CNN）融合像素级和区域级显着性以形成nal显着图，并且联合学习像素级CNN和融合CNN。对四个公共基准数据集的大量定量和定性实验表明，所提出的方法大大优于最先进的显着性检测方法。(A new saliency detection method by significant regional level estimates and forecast significant pixel level and CNN (expressed as CRPSD) combined. For pixel-level significant prediction to perform a full convolution neural network by modifying VGGNet architecture (called pixel-level CNN) learning to perform multi-scale features, image to image prediction to complete the pixel level detection based on the significant learning . For regional levels significantly estimate, the first generation technology to design image is divided into regions based on adaptive super-pixel area, based on the model of the region through the use of CNN (CNN called regional level) to estimate regional levels significantly. By using another CNN (CNN called fusion) Fusion pixel level and regional level to form nal significant saliency map, and the Joint Learning pixel level fusion CNN and CNN. Four common reference data set of a large number of quantitative and qualitative experiments show that the proposed m)

文件列表：

Saliency Detection via Combining Region-Level.pdf,5001460,2017-01-06