The exact visual mechanisms underpinning the approximate number system are still debated. Recent evidence suggests that numerosity is extracted on segmented visual objects, at least for a moderate numerical range (e.g., < 100 items), whereas alternative models rather propose that numerosity is derived from low-level features (e.g., power spectrum) of an unsegmented image, independently from the range. Here, to disentangle these accounts, we generated stimuli that were equalized for spatial frequency amplitude spectrum and luminance across sets of moderate range numerosities (e.g., 9–15 dots), while independently manipulating the perceived item segmentation by connecting dots with illusory contours (ICs). In Experiment 1, participants performed a numerical discrimination task, in which they had to select the numerically larger between two stimuli: a reference stimulus (always 12 dots) and a test stimulus (from 9 to 15 dots) containing 0, 2 or 4 pairs of dots grouped by ICs lines. In Experiment 2, participants were presented only the test stimulus and performed an estimation task. Results clearly showed that in both experiments participants’ performance followed well-known numerical signatures (e.g., distance effect and scalar variability), with numerosity that was underestimated as the illusory connections increased. Crucially, this was found despite spatial frequencies and luminance were kept constant across all the experimental stimuli and these variables were thus uninformative about numerosity. Taken together, these findings indicate that power spectrum in its own cannot explain numerical processing. Rather, visual segmentation mechanisms may be crucial in such processing at least for a moderate numerosity range.

支持近似数字系统的确切视觉机制仍在争论中。最近的证据表明，数量是在分割的视觉对象上提取的，至少在中等数值范围内（例如，< 100 个项目），而替代模型则建议数量来自未分割的低级特征（例如，功率谱）图像，独立于范围。在这里，为了解决这些问题，我们生成了在一组中等范围的数字（例如，9-15 个点）之间针对空间频率幅度谱和亮度进行均衡的刺激，同时通过将点与虚幻轮廓（IC）连接来独立操纵感知项目分割）。在实验 1 中，参与者执行了一项数字判别任务，他们必须在两个刺激之间选择数值较大的一个：一个参考刺激（总是 12 个点）和一个测试刺激（从 9 到 15 个点），包含 0、2 或 4 对按 IC 线分组的点。在实验 2 中，参与者只接受测试刺激并执行评估任务。结果清楚地表明，在两个实验中，参与者的表现都遵循众所周知的数字特征（例如，距离效应和标量可变性），随着虚幻联系的增加，数字被低估了。至关重要的是，尽管空间频率和亮度在所有实验刺激中保持恒定，但这些变量并没有提供关于数量的信息，但这一点仍然存在。综上所述，这些发现表明功率谱本身无法解释数值处理。相当，