Morrone M C, Burr D C, Ross J
Nature. 1983;305(5931):226-8. doi: 10.1038/305226a0.
When a portrait is coarsely quantized into blocks, the block structure hides the face, although lower spatial frequencies of the original image sufficient by themselves for recognition are preserved. Recognition can be recovered by blurring the image, or otherwise attenuating the spurious higher spatial frequency components. Harmon and Julesz claim that high spatial frequencies introduced by quantized blocking mask the lower spatial frequencies which convey information about the face, preventing recognition. Here we show that recognition can be enhanced, without decreasing the amplitude of these spurious higher frequencies, by adding further high-frequency noise to the quantized image. This result is clearly at odds with a theory of high-frequency (or critical band) masking. We suggest that the added noise mutes mechanisms which would otherwise impose a block structure on the image, allowing the alternative perceptual organization of the hidden face to reemerge.
当一幅肖像被粗略地量化为多个块时，尽管原始图像中足以用于识别的较低空间频率得以保留，但块结构会掩盖面部。通过模糊图像或以其他方式减弱虚假的较高空间频率成分，可以恢复识别。哈蒙和朱尔兹声称，量化块引入的高空间频率掩盖了传达面部信息的较低空间频率，从而妨碍了识别。在此我们表明，通过向量化图像添加更多高频噪声，可以增强识别能力，而不会降低这些虚假较高频率的幅度。这一结果显然与高频（或临界带宽）掩蔽理论相悖。我们认为，添加的噪声抑制了那些否则会在图像上强加块结构的机制，使得隐藏面部的另一种感知组织得以重新显现。
