针对前视红外图像质量预测的指标和建模方法进行了研究.通过分析两种典型的目标匹配算法的执行过程,提出了描述图像满足匹配需求的三项准则,即目标均匀性准则、目标局部凸显准则和目标全局凸显准则,进一步提取目标灰度熵、目标背景对比度、目标背景熵差、目标所占面积比例和归一化积相关峰个数这五个图像特征作为质量指标.传统图像质量建模一般采用线性回归方法,但考虑到前视红外图像质量预测模型中二分类因变量的特点,线性回归方法不再适用,因此采用了Logistic回归方法建立模型.测试样本的预测分类表显示,两个模型综合预测准确率均超过80％,证明模型对图像质量具有良好的预测效果.%This paper discussed metrics and modeling of forward looking infrared ( FLIR) image quality prediction. Firstly, it analyzed the implementation process of two typical matching algorithms, followed by presenting of three criteria for describing the requirements from matching algorithms, namely, target uniformity criterion, target local saliency criterion and target global saliency criterion. Furthermore, five image features, segmented target entropy, segmented target to background contrast, segmented target versus background entropy and segmented target pixel ratio and normal cross correlation peak number, were extracted as quality metrics. Linear regression was frequently adopted in traditional image quality modeling, however, considering the binary responses of the FLIR image quality prediction model, it was not suitable to employ linear regression. Therefore , it applied Logistic regression to solve this problem. The prediction classification tables of the test sample show that integrated prediction accuracy of the two models exceed 80% , which demonstrate the effectiveness of the models.
展开▼
