Adaptive Trimmed Mean Filters For Image Restoration

Adaptive Trimmed Mean Filters for Image Restoration
Citation:
A. Restrepo and A. C. Bovik, "Adaptive trimmed mean filters for image restoration," in IEEE
Transactions on Acoustics, Speech, and Signal Processing, vol. 36, no. 8, pp. 1326-1337,
Aug. 1988, doi: 10.1109/29.1660.
Abstract:
The adaptive procedure is based on the selection of an appropriate inner or outer trimmed
mean filter according to local measurements of the tail behaviour of the noise process. The
set of trimmed means used provides robustness against a wide range of noise possibilities
ranging from very shallow tailed to very heavy tailed. A Monte Carlo analysis using a family
of generalized exponential distributions supports the choice of the trimmed mean selected for
measured values of an impulsivity statistic. The assumption underlying the definition of the
filter is that the signal to be filtered is locally smoothly varying, and that the noise process is
uncorrelated and derives from an unknown unimodal symmetric distribution.
Introduction:
Adaptive filters have traditionally been used to overcome the problem of image
nonstationarity. They operate on the premise that, while the image as a whole is
nonstationary, it can be subdivided into small regions, each of which can be treated as
stationary. The size of the small is ill-defined for most applications. As this region is made
smaller, the filter has less information with which to estimate local image activity and may
respond too quickly to noise. If the region is made bigger, the underlying assumptions of
stationarity within the region may be violated. This difficulty is particularly evident in
nonrecursive filters where the region of support must be made explicit. Recursive filters, by
contrast, do not have an explicit region of support.
Causality:
While image filtering is a noncausal task, noncausal recursive filters pose some
implementation difficulties because “future” output values are required to form the current
output value. Because an adaptive (nonlinear) filter is desired, superposition cannot be used
to form the noncausal filter as the sum of a causal and an anticausal filter. If, however, the
filter is to be a variable smoothing filter, then it will curtail filtering activity in the vicinity of
high signal activity. Such a filter does not suffer from causal implementations as causality is
only a significant issue in areas of high signal activity.
Dimensionality:
While the form of a I-D, single pole filter is quite simple, the 2-D recursive filter becomes
significantly more complex. Additional difficulties lie in the fact that, in general, the
denominator of the 2-D transfer function is not factorable. This difficulty can be avoided if
the filter is specified in factored form and the pole locations chosen explicitly.
Stability:
The familiar 2-plane used to determine a recursive filter’s region of convergence in 1-D
becomes a 4-D Z-hyperspace for 2-D recursive filters. The difficulty in predicting filter
stability again stems from the general difficulty in factoring the transfer function. Once again,
if the filter pole locations are explicitly chosen, this problem can be avoided.
Phase:
Recursive filters are, in general, nonzero phase filters. This problem may be further
compounded by the nonlinearities introduced by making the filter adaptive. The phase issue
is avoided in variable smoothing filters that provide no filtering activity in areas of high
signal activity.
Adaptive Recursive Filters For Additive Noise:

The goal of this work is to produce a recursive adaptive image restoration filter that can
provide an estimate of a signal that has been degraded by additive random noise. It is
assumed that the noise signal is formed by a stationary Gaussian process with zero mean and
known variance. Lee’s local statistics filter provides the standard for comparison. Lee’s filter
is a nonrecursive minimum mean squared error linear estimator that provides its estimate
based on the local observation variance
A Recursive Local Wiener Filter:

The original Wiener result was the result of an attempt to optimize squared error in a global
sense. There have been many attempts to produce a local Wiener filter that would produce
locally optimal results in each small neighborhood of pixels. Wiener's optimal filter result is
expressed below
where is the filter's impulse response, Rff(n) is the autocorrelation function of the observed signal,
and Rf,(a) is the crosscorrelation function of the observed signal and the original undegraded signal.
For zero mean, uncorrelated additive noise, this leads to a frequency response of
Emulating The Lee Filter:
Lee's local statistics filter can now be expressed as a variable sized box averager. Once again,
several methods of forming this approximation are available. These include: Minimizing the
squared error between the Lee step response and the recursive step response; Ensuring ninety
percent of the step response takes the same duration
Conclusion:
It has been demonstrated that it is possible to create 2-D adaptive recursive filters without
being significantly impaired by issues of phase distortion, causality, and stability. This
adaptive recursive filtering framework is used to address the problem of restoring images
degraded by additive Gaussian noise. Three approaches to providing filters that are optimal in
the squared error sense are explored. The results obtained using the schemes outlined
compare favorably with the many adaptive nonrecursive solutions. The adaptive implicit
region of support offered by the adaptive recursive filter yields important improvements over
nonrecursive results and offers significant potential for effective and efficient restoration and
enhancement filtering.
References:
G. L. Anderson and A. N. Netravali, “Image restoration based on a subjective criterion,”
IEEE Trans. Syst., Man, Cybem., vol. SMC-6, no. 12, pp. 845-853, Dec. 1976. [2] M. G.
Bellanger, Adaptive Digital Filters and Signal Analysis. New York: Marcel Dekker, 1987.K.
J. Erler, “Adaptive recursive image filtering,” M.S. thesis, University of Waterloo, Waterloo,
Ontario, Canada, 1991.V. S. Frost et al., “A model for radar images and its application to
adaptive digital filtering of multiplicative noise,” IEEE Trans. Pattern Anal., Machine Intell.,
vol. PAMI-4, no. 2, pp. 157-166, Mar. 1982. the introduction of VLSI and rapid development
of microprocessor circuits, digital filter applications are increasing at an expanding rate In
general, the design of 2-D digital filters usually involves two related phases. In the first
phase, the mathematical representation of a filter is obtained from a set of specifications or
given data either in state-space or transfer-function form (approximation phase). In this
phase, it is particularly desirable to obtain a simplified model that greatly reduces the
computation effort or the hardware complexity of the filter. On the other hand, given a
transfer function, there are theoretically an infinite number of realizations to implement such
a filter (implementation phase).

Adaptive Trimmed Mean Filters For Image Restoration

Uploaded by

Copyright:

Available Formats

Adaptive Trimmed Mean Filters For Image Restoration

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Adaptive Trimmed Mean Filters For Image Restoration

Uploaded by

Copyright:

Available Formats

Adaptive Trimmed Mean Filters for Image Restoration

Adaptive Recursive Filters For Additive Noise:

A Recursive Local Wiener Filter:

You might also like