Added noise restores recognizability of coarse quantized images

Abstract

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.