基于视觉注意力模型的红外人体图像兴趣区域提取

Extraction of region of interest in FIR pedestrian based on human visual attention model

由于红外人体图像所处背景的复杂性,使用阈值分割定位红外人体图像往往把大量的非人体图像一起分离出来,这给后续工作带来了巨大的计算量。研究了红外人体图像的典型特点和人眼视觉判断人体轮廓图像的敏感特征,使用该特征建立了简单的人眼视觉对人体的注意力模型,并构建特征判别概率函数,对待定目标进行概率的估算,最终使用该模型快速地剔除非人体红外图像的干扰,并准确地定位人体兴趣区域。

Because of the FIR pedestrian under the condition of complex background, threshold segmenation method for infrared images always extract region of interest that embodys non-human infrared image, which bring huge workload to follow-up work. This paper studied the infrared human body image features and human visual attention to human body, constructed human visual attention model to human body, constructed probability function to estimate FIR pedestrian. Finally achieved the object that eliminate of interference non-human infrared image, accurate position region of interest of FIR pedestrian.