摘要
提出了一种对图像单目标识别算法性能进行量化评估的方法。该方法将ROC曲线予以量化,在高斯分布假设下推导出识别算法性能的单一量化指标评估模型,采用均值和方差综合描述算法效果。并采用Sigmoid函数对Gaussian分布函数逼近的方法,结合Logistic回归分析模型,推导了评估模型的简化解算公式。该方法实现了对识别算法效果的定量评估,为不同应用环境下优化算法参数和算法组合提供了量化依据。仿真结果表明了该方法的有效性。
A quantitative performance evaluation method for image target detection is proposed. The Receiver Operating Characteristic （ROC） curve is quantified, and based on Gaussian distributing model, a single quantitative evaluation model is established, and the mean and standard deviation are employed to compute the performance value. The evaluation model is further simplified and becomes feasible by using Sigmoid function to approach to Gaussian distributing model, and employing Logistic model to analyze and compute the function. The quantitative performance evaluation is carried out, providing a quantitative foundation for optimizing the parameters and combination of the algorithms in different applications. Test results show the effectiveness of this method.
出处
《半导体光电》
CAS
CSCD
北大核心
2009年第5期788-792,共5页
Semiconductor Optoelectronics
