Zhou Wang, 12/2/03
<br>
<br>

Blur Perception and Local Phase Statistics of Natural Images
<br>
<br>
<br>

It is commonly assumed that our visual systems detect blur of visual images
by sensing the reduction of energy at high spatial frequencies. We argue
that the disruption of local phase structure is a more important factor in
blur perception. We demonstrate that a sharp image with its high frequency
energy reduced but local phase preserved appears much sharper than a blurred
image with its high frequency energy corrected but local phase unchanged. We
show that precisely localized isolated features such as edges, lines and
points result in strong structural regularity of local phase across spatial
scale and position. We develop a cross-scale phase prediction technique in
the complex wavelet transform domain and use it to measure the structural
regularity of local phase. Experiments on 1000 nautral images and their
blurred versions show that this property is significant in natural images
and is substantially disrupted by blurring. This study not only provides a
new perspective on blur perception and local image structures, but also
suggests a mechanism for "seeing beyond the Nyquist limit". It has a broad
range of potential applications in machine vision, image processing, and the
design of super-precision signal detection devices.
<br>
<br>

http://www.cns.nyu.edu/~zwang