Browse Prior Art Database

ADAPTIVE OPTIMAL LINEAR FILTERING FOR 1-TO-N BIT HALFTONE DESCREENING

IP.com Disclosure Number: IPCOM000026876D
Original Publication Date: 1994-Feb-28
Included in the Prior Art Database: 2004-Apr-06
Document File: 4 page(s) / 227K

Publishing Venue

Xerox Disclosure Journal

Abstract

When processing a digital image it is desirable to have several bits of gray scale resolution per pixel. The processing may be any of a number of typical operations such as tone rendition modification, halftoning with a screen optimized for a particular process, edge sharpening, or resolution conversion. Often a document to be processed, such as a halftoned image, is available only in binary, one bit per pixel, form. Prior to processing, conversion to multibit quantization is currently being accomplished in Xerox@ scanners (e.g.,7650 Pro Imager) by using low-pass filtering techniques to remove the halftone screen and give an estimate of a multibit per pixel version of the original image. This type of filtering tends to blur edges and lose fine detail. One technique employs Logical Filtering and offers substantially better results than traditional low-pass filtering. The present disclosure describes an adaptive filtering technique for l-to-N bit conversion of halftone images that performs a more accurate reconstruction than low-pass filtering and is statistically optimal in a mean square sense for a collection of images.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 42% of the total text.

Page 1 of 4

XEROX DISCLOSURE JOURNAL

ADAPTIVE OPTIMAL LINEAR Proposed Classification FILTERING FOR 1-TO-N BIT
HALFTONE DESCREENING
Robert P. Loce
Jeffrey D. Kingsley
James J. Appel

U.S. C1.358/456 Int. C1. H04n 1/40

When processing a digital image it is desirable to have several bits of gray scale resolution per pixel. The processing may be any of a number of typical operations such as tone rendition modification, halftoning with a screen optimized for a particular process, edge sharpening, or resolution conversion. Often a document to be processed, such as a halftoned image, is available only in binary, one bit per pixel, form. Prior to processing, conversion to multibit quantization is currently being accomplished in Xerox@ scanners (e.g.,7650 Pro Imager) by using low-pass filtering techniques to remove the halftone screen and give an estimate of a multibit per pixel version of the original image. This type of filtering tends to blur edges and lose fine detail. One technique employs Logical Filtering and offers substantially better results than traditional low-pass filtering. The present disclosure describes an adaptive filtering technique for l-to-N bit conversion of halftone images that performs a more accurate reconstruction than low-pass filtering and is statistically optimal in a mean square sense for a collection of images.

Halftoning, which is a thresholding process, causes a loss in information. Descreening is an attempt to restore, or best estimate, the original image. Low-pass filter descreening does not use knowledge of the specific screening process used to create the binary image although this knowledge will allow for a better restoration. Thus the first step in the proposed method of adaptive descreening is to obtain this information.

The halftone screen information may be obtained in one of several ways. The most simple method may be to have the threshold matrix stored or passed with the image. Another method involves the use of autocorrelation and Hough transform techniques. A third method, proposed here, uses Fourier analysis and statistical properties of the cells. In this method the distance of the largest magnitude frequency component of the Fourier transform to the origin of the transform gives the screen frequency. The angle of this component with respect to the axes gives the screen angle. With this information the binary halftone image can be divided into halftone cells. The statistical properties of the cells can then be used to determine the filling order and estimate the thresholds. For example, all the cells containing only one black pixel can be examined to find the location of that pixel within each cell. The location which occurs most frequently has the lowest threshold (turned black at lowest density). Next, all cells with only two black pixels would be examined to determine which pixel location in the cell is most likely to be second in the fill-

XEROX DISCLOSURE JOURNAL - Vol. 19, No. 1 JanuaryPebruary 1994...