基于高层先验语义的显著目标检测

Salient target detection based on high-level priori semantics

为解决图像低级特征不能够均匀进行显著目标检测的问题,将高层先验语义和低级特征进行结合,提出一种新颖的基于高层先验语义的显著目标检测算法模型。利用深度卷积神经网络对输入图像以及显式显著性先验信息分别进行语义分割提取,得到显式显著性检测图;通过将图像中隐含的先验显著性特征与显著性值进行映射得到训练模型计算隐式显著性图;将显式显著性检测图和隐式显著性检测图进行自适应融合,形成均匀覆盖显著目标像素的精确显著检测图。为验证算法模型的有效性,将算法在具有挑战性的ECSSD和DUT-OMRON图像数据集进行实验仿真,实验结果表明,该算法的显著目标检测效果较其他方法有较为显著的提升。

In order to solve the problem that the low-level features of the image can not uniformly detect the saliency target,this paper combines the high-level a priori semantics with the low-level features,and proposes a novel algorithm based on high-level priori semantics.Firstly,the deep convolutional neural network is used to extract the input image by semantic segmentation,and the explicit saliency prior information is semantically segmented and extracted to obtain the explicit saliency detection map.By mapping the prior significance feature and the significance value in the image,the training model is used to calculate the implicit significance map;the explicit significance map and the implicit significance map are adaptively fused to form a saliency detection map that uniformly covers the saliency target pixels.In order to verify the validity of the algorithm model,the algorithm is simulated in the challenging ECSSD and DUT-OMRON image datasets.The extensive experimental results show that the proposed algorithm has a significant saliency improvement in the target detection effect compared with other methods.