Guyader Nathalie, Chauvin Alan, Boucart Muriel, Peyrin Carole
Univ. Grenoble Alpes, GIPSA-Lab, F-38000 Grenoble, France; CNRS, GIPSA-Lab UMR5216, F-38000 Grenoble, France.
Univ. Grenoble Alpes, LPNC, F-38000 Grenoble, France; CNRS, LPNC UMR5105, F-38000 Grenoble, France.
Vision Res. 2017 Apr;133:100-111. doi: 10.1016/j.visres.2016.12.019. Epub 2017 Mar 30.
The visual perception of human faces by man is fast and efficient compared to that of other categories of objects. Using a saccadic choice task, recent studies showed that participants were able to initiate fast reliable saccades in just 100-110ms toward an image of a human face, when this was presented alongside another image without a face. This extremely fast saccadic reaction time is barely predicted using classical models of visual perception. Thus, the present research investigates whether this result might be explained by the low spatial frequency content of images. Using the same paradigm, with two images simultaneously presented to the left or right visual fields, participants were asked to make a saccade towards a target image. The target was defined as an image belonging to one category: human face, animal or vehicle. The other image corresponded to the distractor and belongs to the other categories. We compared performance to saccade toward one category of target. The two images were displayed either in color, gray-level, low-pass filtered or high-pass filtered. As previous studies, we found that the shortest SRT was observed for saccades towards faces rather than towards animals or vehicles. Analysis of saccadic reaction time distributions showed that, in 130-140ms, participants were able to make more correct than incorrect saccades towards faces for unfiltered (color and gray-level) and low-pass filtered images whereas they needed more time for high-pass filtered images. In contrast, the minimum time participants needed to correctly saccade towards animals and vehicles was longer for low-pass and high-pass filtered than for unfiltered images. The analysis of the image statistics in the Fourier domain revealed that the amplitude spectrum of faces was mainly contained in the low spatial frequencies. Consistent with a coarse-to-fine processing of visual information, our results suggest that extremely fast saccades towards faces could be initiated by low spatial frequencies.
与其他类别物体相比,人类对人脸的视觉感知快速且高效。近期研究通过扫视选择任务表明,当人脸图像与另一无脸图像同时呈现时,参与者能够在短短100 - 110毫秒内朝着人脸图像发起快速可靠的扫视。使用传统视觉感知模型几乎无法预测这种极快的扫视反应时间。因此,本研究探讨这一结果是否可由图像的低空间频率内容来解释。采用相同范式,同时向左右视野呈现两幅图像,要求参与者朝着目标图像进行扫视。目标被定义为属于某一类别(人脸、动物或车辆)的图像,另一图像为干扰物,属于其他类别。我们比较了朝着一类目标进行扫视的表现。这两幅图像分别以彩色、灰度、低通滤波或高通滤波形式显示。如先前研究一样,我们发现朝着人脸的扫视观察到的最短扫视反应时间,而非朝着动物或车辆的扫视。对扫视反应时间分布的分析表明,在130 - 140毫秒内,对于未滤波(彩色和灰度)及低通滤波图像,参与者朝着人脸进行的正确扫视多于错误扫视,而对于高通滤波图像则需要更多时间。相比之下,对于低通和高通滤波图像,参与者朝着动物和车辆正确扫视所需的最短时间比未滤波图像更长。在傅里叶域对图像统计的分析表明,人脸的幅度谱主要包含在低空间频率中。与视觉信息从粗略到精细的处理一致,我们的结果表明朝着人脸的极快速扫视可能由低空间频率引发。