Similarity of Low-Pass Filtered and Pixelated Images at Different Time Scales

摘要

In digital image editing, users perform editing operations on rasterized images. Depending on the resolution of their screen, these rasterized images appear to be pixelated, that is, individual pixels of the image are perceived as discrete units, thereby changing its spatial frequency spectrum. Changes in spatial frequency spectra on perception are usually investigated by low-/high-pass filtering. We studied to what degree are low-pass filtered images perceptually similar to pixelized images, in explicit (Experiment-1) and implicit (Experiment-2) tasks. For every base-image, we created one pixelized version (pixel-size corresponded to 1 deg of visual angle) and six levels of low-pass filtered images. For Experiment-1, 12 participants chose from six low-pass filtered images with increasing levels of blur, the one that appeared most similar to the pixelized version of the same image. The trials were not timed and 100 images from four categories (landscape with/without water, macro with humanon-human focal object) were used. For (long, unconstrained) inspection durations tested in Experiment-1, participants chose the least blurry image as most similar to the pixelized version. In Experiment-2, with 2IFC-task, 14 participants discriminated between low-pass filtered (three-levels) and pixelized versions of the same image, for three time-intervals (40ms, 70ms, 100ms). Discrimination performance was average over interval durations at 82.42% for the pixelized and the original base-image. It plateaued at about 70% for medium low-pass filtered and pixelated version. Accuracy increased with prolonged interval durations. For (short) inspection durations tested in Experiment-2, the pixelized image was perceptually most similar to the versions with greater blur. We conclude that results from studies of low-pass filtered images can be generalized to pixelized version for durations less than 100ms but not for the longer inspection durations that are usually used in digital photo-editing processes. Thus new studies with pixelized-images are required to study the perceptual process of digital photo-editing.