基于视觉注意机制的小目标预检测特性

Performance analysis of small target pre-detection based on visual attention mechanism

针对一种使用了离散非线性采样的视觉注意机制模型,对其应用在小目标预检测上的检测概率进行了研究和分析.在一定的假设条件下,从理论上推导出了模型中影响小目标预检测概率的各个因素,包括目标大小、目标所在采样节点的大小、输入图像信噪比和目标均匀性参数等.由于采用了非线性的采样模型,注视点的转移将带来不同图像区域空间分辨率的变化,从而不能获得总的检测概率与上述各参数的有意义的关系,通过实验分析了这些因素在小目标预检测中对检测概率的具体影响.

A visual attention mechanism model for discrete non-linear sampling is studied to analyze its detection probability during pre-detecting small targets. Based on some reasonable hypothesizes, the factors influencing the detection probability were deduced, including the target size, the size of the sampling node which contains the target, the signal-noise ratio of the input image and the uniformity of the target. As a result of the non-linear sampling model and change of the space resolution in different regions of the image brought by the transition of the gaze point, a specific relationship between the overall detection probability and the various factors could not be deduced, some experiments were performed to find the detailed effects of these factors on the de.tection probability.