📄 中文摘要
面部比较的准确性并非静态整数，而是由匹配阈值控制的动态权衡。当供应商声称“99%准确率”时，实际上是指在理想照明条件下、使用高分辨率正面图像测得的接收者操作特征（ROC）曲线上的一个单一点。在实际应用中，从护照风格照片转向“野外”图像（如带有运动模糊或角度偏差的闭路电视画面）可能会使性能下降多达40%。理解这些系统的实际表现需要关注假阳性率（FMR）与假阴性率（FNMR）之间的数学权衡。
面部识别准确性是如何真正测量的——以及这为何重要
📄 English Summary
How Facial Recognition Accuracy Is Really Measured — And Why It Matters
Facial comparison accuracy is not a static integer but a dynamic trade-off controlled by the match threshold. When vendors claim '99% accuracy,' they refer to a single point on a Receiver Operating Characteristic (ROC) curve, likely measured under ideal lighting conditions with high-resolution frontal images. In real-world scenarios, transitioning from passport-style photos to 'wild' imagery—such as CCTV frames with motion blur or off-angle poses—can degrade performance by as much as 40%. Understanding the actual performance of these systems requires attention to the mathematical trade-off between False Match Rate (FMR) and False Non-Match Rate (FNMR).