基于贝叶斯推理的多线索视觉注意模型

摘要

针对大多数视觉注意模型都采用简单加权线性融合的方式获取显著图,提出了一个更符合生物学机制的基于贝叶斯推理的多线索视觉注意模型,模拟视觉系统腹侧通路与背侧通路中的视觉注意过程,采用贝叶斯推理的方式集成自顶向下与自底向上的信息,同时还集成了多种视觉线索,包括形状、颜色和上下文等.利用该模型进行遥感影像中的目标检测与定位的结果表明,该模型能有效的检测出目标并给出目标所在的位置.%For most of the visual attention models use linear combination to obtain saliency map,a visual attention model based on Bayesian inference using multiple cues is proposed,which is more consistent with biological mechanism.Bayesian inference is used to combine both top-down and bottom-up information,and the processes of visual attention is simulated in ventral stream and dorsal stream.Multiple visual cues are integrated,such as shape,color,and context.The experimental results on object detection and localization in remote sensing images show this model can detect and locate objects effectively.