Rapid categorization of achromatic natural scenes:: how robust at very low contrasts?
Rapid categorization of achromatic natural scenes:: how robust at very low contrasts?
复制标题
DOI：
10.1111/j.1460-9568.2005.04029.x
复制
发表时间：
2005-04-01
影响因子：
3.4
通讯作者：
Fabre-Thorpe, M
中科院分区：
文献类型：
--
作者：
Macé, MJM;Thorpe, SJ;Fabre-Thorpe, M
人类的视觉系统非常擅长对物体进行分类，即使在具有挑战性的视觉条件下。在这里，我们专门评估了鲁棒性的视觉系统在面对大的对比度变化，在一个高层次的分类任务，使用自然图像。人类受试者进行了一个去/不去动物/非动物分类任务，短暂闪烁的灰度图像。对随机提供给受试者的大范围对比度条件下的性能进行了分析，从正常到初始对比度的3%不等。准确度非常稳健，受试者的表现远高于机会水平（大约70%正确），只有10-12%的初始对比度。准确度降低，对比度降低，但达到机会水平，只有在最极端的条件下（初始对比度的3%）。相反，平均反应时间的最大增加约为60 ms（初始对比度的8%）;然后随着对比度的进一步降低而保持稳定。在正确目标和干扰试验中记录的相关ERP表现出明显的差异效应，其幅度和峰潜伏期分别与任务准确性和平均反应时相关。这些数据表明，在非常低的对比度下，视觉系统在对象分类中具有很强的鲁棒性。他们认为大细胞信息可以在腹侧流视觉功能（例如物体识别）中发挥作用。表现可能依赖于早期的对象表示，缺乏随后由小细胞系统提供的细节，但包含足够的信息，以达到在分类任务的决定。
The human visual system is remarkably good at categorizing objects even in challenging visual conditions. Here we specifically assessed the robustness of the visual system in the face of large contrast variations in a high-level categorization task using natural images. Human subjects performed a go/no-go animal/nonanimal categorization task with briefly flashed grey level images. Performance was analysed for a large range of contrast conditions randomly presented to the subjects and varying from normal to 3% of initial contrast. Accuracy was very robust and subjects were performing well above chance level (approximate to 70% correct) with only 10-12% of initial contrast. Accuracy decreased with contrast reduction but reached chance level only in the most extreme condition (3% of initial contrast). Conversely, the maximal increase in mean reaction time was approximate to 60 ms (at 8% of initial contrast); it then remained stable with further contrast reductions. Associated ERPs recorded on correct target and distractor trials showed a clear differential effect whose amplitude and peak latency were correlated respectively with task accuracy and mean reaction times. These data show the strong robustness of the visual system in object categorization at very low contrast. They suggest that magnocellular information could play a role in ventral stream visual functions such as object recognition. Performance may rely on early object representations which lack the details provided subsequently by the parvocellular system but contain enough information to reach decision in the categorization task.