摘要
研究采用FaceReader 7.0版本,对国内常见面部表情图片库进行了识别分析。结果发现:(1)软件对USTC-NVIE图片库420张人为表情图片的识别率为58.0%(其中不戴眼镜的面部表情图片识别率可达64.0%,戴眼镜的图片识别率为51.9%);(2)对CFAPS图片库840张图片的识别率为48.1%。结论:软件对国内常见面部表情图片库的识别有效性尚可,对"高兴"、"惊奇"和"中性"情绪表现出较好的识别率,但其对"生气"、"害怕"以及戴眼镜面部表情图片的识别率还有待进一步提高。因此相关研究者在进行较为重要或精确的研究时应该注意到这一问题,必要时进行人工校对,以免影响研究结果。
FaceReader 7. 0 was applied to recognize the facial expression images in two common expression database sys-tems in China, USTC-NYIE and CFAPS. The results showed that： （1） the recognition rate to USTC-NYIE reached 58. 0% , in which the facial images without glasses recognition rate is up to 64. 0 % while facial images wearing glasses recognition rate is 51. 9%. （2） the recognition rate to CFAPS was 48. 1%. In a word, the recogni-tion validity of FaceReader is good to recognize Chinese common facial image database systems, and the recognition rate to the emotions “ happy ”,“ surprise” and “ neutral” is higher. However, the recognition rate to “ angry”, “ fear” and facial images wearing glasses remains to be further improved. Therefore, researchers should pay attention to this problem in using the software to finish important or precise studies, and manual proofreading should be added to improve the results of the study if it’s necessary.
出处
《心理技术与应用》
2018年第2期100-108,共9页
Psychology(Techniques and Applications)
基金
江苏高校哲学社会科学研究基金项目(2015SJB122)
江苏高校自然科学研究面上项目(17KJD190002)
南京师范大学哲学社会科学青年科研人才培育基金项目(14QNPY09)
