基于颜色与梯度布尔特征融合的图像显著性检测

Saliency Detection Based on Color and Gradient Boolean Features

图像显著性检测存在区域不均匀、显著性值低的问题,本文在对BMS(Boolean Map based Saliency)模型进行研究的基础上提出了基于颜色与梯度布尔特征融合的显著性检测模型(Boolean Map of Color and Gradient based Saliency,BMCG)。根据Gestalt前背景分离的原则,通过随机阈值化颜色通道和梯度通道产生含有图像拓扑结构的二进制布尔新息图,进一步生成视觉注意图并进行线性融合,经过后处理形成显著性图。仿真结果表明BMCG算法比BMS算法的召回率提高了2.12%,准确率提高了4.56%。

There exist non-uniform areas and low saliency score problems in image saliency detection.By analyzing the model of Boolean map based saliency,this paper proposes the Boolean maps of color and gradient based saliency model（BMCG）.According to Gestalt principle of figure-ground segregation,the Boolean maps with the topological structure are generated via the random threshold color channels and gradient channel.Furthermore,these Boolean maps are refined into the attention maps of visual and are linearly combined to generate the saliency map via post-process.The simulation results show that BMCG algorithm is better than BMS algorithm,improving the recall rate of 2.12% and precision rate of 4.56%.