Low-level image properties in facial expressions
Low-level image properties in facial expressions
复制标题
DOI：
10.1016/j.actpsy.2018.05.012
复制
发表时间：
2018-07-01
影响因子：
1.8
通讯作者：
Hayn-Leichsenring, Gregor U.
中科院分区：
文献类型：
--
作者：
Menzel, Claudia;Redies, Christoph;Hayn-Leichsenring, Gregor U.
我们研究了人脸照片的低层图像属性，并分析了它们是否会随着个体所表现出的不同情绪而变化。在三个数据库中测量了图像属性的差异，这三个数据库总共描绘了167个人。脸部图像要么以原始形式使用，要么被剪裁成标准格式，要么与面具叠加。分析的图像特性有：亮度、红度、黄度、对比度、光谱斜率、总功率和低、中、高空间频率的相对功率。结果表明，图像属性在每个单独图像集中的不同表情之间存在显著差异。此外，特定的面部表情与所有三个数据库中一致的图像属性模式相对应。为了从实验上验证我们的发现，我们均衡了三幅图像的亮度直方图和光谱斜率，这些图像来自一个特定的人，他们表现出两种表情。与原始图像三联体相比，参与者在平衡图像三合一中匹配表情的速度明显较慢。因此，这些图像属性(即，光谱斜率、亮度或对比度)中的现有差异有助于在特定脸部图像集合中进行情感检测。
We studied low-level image properties of face photographs and analyzed whether they change with different emotional expressions displayed by an individual. Differences in image properties were measured in three databases that depicted a total of 167 individuals. Face images were used either in their original form, cut to a standard format or superimposed with a mask. Image properties analyzed were: brightness, redness, yellowness, contrast, spectral slope, overall power and relative power in low, medium and high spatial frequencies. Results showed that image properties differed significantly between expressions within each individual image set. Further, specific facial expressions corresponded to patterns of image properties that were consistent across all three databases. In order to experimentally validate our findings, we equalized the luminance histograms and spectral slopes of three images from a given individual who showed two expressions. Participants were significantly slower in matching the expression in an equalized compared to an original image triad. Thus, existing differences in these image properties (i.e., spectral slope, brightness or contrast) facilitate emotion detection in particular sets of face images.