摘要： 在评价人脸识别算法或系统性能时，通常采用识别率来衡量识别的效果。但识别率仅表示识别通过的数量比率，并不能表示识别通过的质量，不能准确地反映出识别的效果。该文提出了一种基于危险系数的性能评价方法，能很好地解决这个问题。通过红外人脸识别的实验，验证该评价方法的可行性。这种评价方法还可以被推广到目标识别的其它领域。

关键词: 人脸识别；性能评价；危险系数；识别率

Abstract: Recognition rate is often used to measure the effectivity of face recognition algorithms or systems when evaluating their performance. But recognition rate only shows the quantity of correct recognition, not the quality. So, it can’t measure the effectivity of recognition accurately. This paper presents a performance evaluation method based on danger coefficient, which can solve the problem well. The results of experiment on infrared face recognition show this method is feasible. The method can also be employed in other fields of object recognition.

Key words: Face recognition; Performance evaluation; Danger coefficient; Recognition rate