Weber’sLawwas first described in 1834 by German physiologist E. H. Weber [Web34],
and was later formulated quantitatively by the great experimental psychologist Gustav
Fechner [Fec58], founder of the modern psychophysics. The law reveals the universal
influence of the background stimulus on human’s sensitivity to the intensity incre-
ment , or the so called JND (just-noticeable-difference), in the perception of both
sound and light. It claims that the so-called Weber’s fraction is a constant:
const. (1)
Many experiments have demonstrated that in a large range of stimuli , Weber’s Law
indeed provides a good approximation.
Empirical evidence is easy to gather from daily life for a qualitative understanding
of Weber’s Law. In a fully packed stadium where the background sound intensity is
high, one has to speak close to shouting in order to be effectively heard by the other
folks. The same observation is true for visual communication. The stars can be clearly
spotted in a dark night without a bright full moon, and away from the urban neon
lights. But otherwise our naked vision has much difficulty in finding them. In the
current paper, we apply Weber’s Law in the context of visual perception. Therefore,
stands for the background light intensity and the intensity fluctuation.
Since almost all images are eventually to be observed and interpreted by humans,
an ideal digital image processor has to take into account the effects of human psychol-
ogy and psychophysics, such as that of Weber’s Law. This is an important new area that
needs to be further explored. The current paper makes the first infantile attempt of in-
tegrating Weber’s Law into image restoration schemes. We demonstrate our main idea
through “Weberizing” the well known classical model of total variation (TV) denoising
and enhancement by Rudin, Osher, and Fatemi [ROF92, RO94].
The organization goes as follows. In Section 2, we quickly review the TV restora-
tion model in image processing, and explain the idea behind its “Weberization.” In
Section 3, we first rigorously interpret the Weberized TV restoration energy and its ad-
missible space, and then apply the direct method to study the existence and uniqueness
of the minimizers. The computational approach to the minimization of the Weberized
TV energy is addressed in Section 4, accompanied by some typical numerical results.
The conclusion goes into Section 5.
2 Weberized TV for Image Restoration
Let denote the observed raw image data, which is assumed to be a degraded version
of the original good image . Distortions in are typically modeled by blurring and
noising:
(2)
where is a linear blurring operator, or a lowpass filter with , and denotes
white noise. The goal of image restoration is to recover the original good image
from one single observation of (since strictly speaking, is a random field). In
this paper, we shall assume that the noise is spatially homogeneous, and can be well
2