Image analysis – Image enhancement or restoration – Image filter
Reexamination Certificate
2000-04-27
2004-11-09
Mehta, Bhavesh M. (Department: 2625)
Image analysis
Image enhancement or restoration
Image filter
C708S300000, C708S314000, C708S819000
Reexamination Certificate
active
06816621
ABSTRACT:
FIELD OF THE INVENTION
The present invention is generally related to the field of signal processing and, more particularly, to the field of programmable hardware and/or software filtering of image data in compact architectures with windowed shift-invariant filters operating on signals with two or more levels per sample.
BACKGROUND OF THE INVENTION
In the arts it is often necessary to manipulate a received signal or image, for both practical and operational reasons. For instance, the signal may need to be compressed prior to some secondary operation being performed on the data such as transmission over a cable or phone wire. On the receiving end, compressed transmitted signals may eventually need to be restored or reconstructed to return the signal upon receipt to its original form. Other operations can take the form of feature detection or pattern recognition, among others. In order to manipulate the received signal or image data in these instances, mathematical operations are often performed. To achieve this, the discrete signal or image is often converted into or represented by a mathematical sequence or array of discrete values. For example, binary signals or images often take the form of discrete values of [0,1] and 8-bit gray-scale images often take the form of values in the range of [0 . . . 255]. The field of discrete signal and image processing, both theoretical and applied, is relatively dominated by mathematics in the form of specific methods and applications.
Many methods of signal or image processing use one or more applications involving filters. A filter operates by capturing a windowed observation of an image or signal. A windowed observation is a collection or vector of samples from the captured image or signal. Based on this vector, a decision, based on some formula or algorithm, is made about the image or signal at a particular position within the image or signal. In general, a filter operates on a set of signal or image samples which are mathematically written as a vector having the form of (X
1
, . . . , X
N
).
Many techniques can be found in the arts, of which the following are representative and which are hereby incorporated by reference being made thereto. For example, a filter could be employed to assign one or more specific code-words to a given sample as shown in
FIG. 1
, (often called vector quantization), as taught by
Vector Quantization and Signal Compression,
A. Gesho and R. M. Gray, 1991. A filter can also assign one or more tags to a center pixel in order to indicate that the vector represented pixels are part of a specific feature of the captured image, such as a line, letter, or symbol. It can also assign a new sample value to the center pixel in order to accomplish the overall goal of a restoration or enhancement of a degraded image, as taught by
Enhancement and Restoration of Digital Documents,
R. P. Loce and E. R. Dougherty, SPIE Press, 1997, and
Enhancement of Digital Documents,
R. P. Loce and E. R. Dougherty, Electronic Imaging Technology, SPIE Press 1999, and
Two
-
Dimensional Signal and Image Processing,
J. S. Lim, Prentice Hall, 1990. A filter can also be used to predict a sample or pixel value outside the capture window with the aim of reducing the amount of data needed to store the image. This technique is often referred to as predictive encoding as taught by
Digital Image Compression Techniques,
M. Rabbani and P. W. Jones, SPIE Press, 1991. A filter can be used to assign an array of fewer, more restrictive values to this observation (often referred to as either “quantization,” “thresholding,” or “dithering” as taught by
Digital Halftoning for Printing and Display of Electronic Images,
R. P. Loce, P. G. Roetling, and Y. W. Lin, Electronic Imaging Technology, SPIE Press, 1999. Many others are found in the known literature. Furthermore, other applications of filters used in signal or image processing include, resolution conversion, object detection, speckle-removal, and edge enhancement.
Nonlinear image or signal processing is a general representation of signal or image filtering based on a logical decomposition of a filter into a set of relatively simple operators. Any windowed shift-invariant filter can be represented as a combination of simple operations called hit-or-miss transforms as taught by
Nonlinear Filters for Image Processing,
E. R. Dougherty and J. Astola (eds.), SPIE/IEEE Press, 1999. The term nonlinear refers to the entire class of windowed filters that includes linear filters, i.e., those filters computed by convolution.
Since filters can be highly complex and repeatedly be executed millions of times in rapid succession in order to perform a single operation on a given image or signal, what is needed in the arts are compact, efficient computing architectures directed towards this end. Efficient architectures depend on filter representations. If a filter can be broken down into its base operations, it might be possible to execute the relatively simple operations in parallel in order to achieve high-speed execution and overall efficiency.
Computational Gray
-
Scale Morphology on Lattices
(
A Comparator
-
Based Image Algebra
)
Part I: Architecture,
E. R. Dougherty and D. Sinha, Real Time Imaging, Vol. 1, No. 1, 1995, teaches a computer architecture using representations that perform tests in parallel. The representational methods as taught therein express any filtering operation as a collection of logical tests or operations wherein each logical test is associated with a value. If a windowed observation satisfies a test, the filter output is then associated with that value. Although the above architecture is not practical for filters requiring many tests, it does teach a general representational method.
It is the aim of the present invention to overcome the conventional prior art print limitations, as described above, by performing logical comparisons in a more compact and efficient way.
BRIEF SUMMARY OF THE INVENTION
A method for implementing a filter for processing a discrete signal having the steps of first obtaining a plurality of sample values from the discrete signal; then using each sample value to retrieve a bit vector from a plurality of tables. Afterwhich, a logical AND step is performed on the set of retrieved bit vectors; and a determination is made as to whether the resultant bit vector is comprised of all zeros or not. If so, then a predetermined default action is performed, otherwise the position of a non-zero bit in the bit vector is determined and the non-zero bit position is then used to generate a value of the filter.
REFERENCES:
patent: 4589084 (1986-05-01), Fling et al.
patent: 4661948 (1987-04-01), Shapiro et al.
patent: 4771396 (1988-09-01), South et al.
patent: 4941191 (1990-07-01), Miller et al.
patent: 5689592 (1997-11-01), Ericsson et al.
patent: 5873081 (1999-02-01), Harel
patent: 6071004 (2000-06-01), Le Gall et al.
patent: 6173388 (2001-01-01), Abercrombie et al.
Fay Sharpe Fagan Minnich & McKee LLP
Mehta Bhavesh M.
Tabatabai Abolfazl
Xerox Corporation
LandOfFree
Method for generating shift-invariant filters does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for generating shift-invariant filters, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for generating shift-invariant filters will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3339983