Image analysis – Pattern recognition – Classification
Reexamination Certificate
1998-07-07
2001-02-06
Lee, Thomas D. (Department: 2724)
Image analysis
Pattern recognition
Classification
C382S168000, C382S172000, C358S451000, C358S534000
Reexamination Certificate
active
06185335
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Technical Field
The invention relates to image processing. More particularly, the invention relates to image classification and halftone detection, especially with regard to digitized documents, acquired for example by digital scanning, and the reproduction of such images on digital color printers.
2. Description of the Prior Art
Electronic documents contain a variety of information types in various formats. A typical page of such document might contain both text (i.e. textual information) and images (i.e. image information). These various types of information are displayed and reproduced in accordance with a particular formatting scheme, where such formatting scheme provides a particular appearance and resolution as is appropriate for such information and printing device. For example, text may be reproduced from a resident font set and images may be reproduced as continuous tone (contone) or halftone representations. In cases where a halftone is used, information about the specific screen and its characteristics (such lines per inch (Ipi)) is also important.
It is desirable to process each type of information in the most appropriate manner, both in terms of processing efficiency and in terms of reproduction resolution. It is therefore useful to be able to identify the various information formats within each page of a document. For example, it is desirable to identify halftone portions of a document and, as appropriate, descreen the halftone information to provide a more aesthetically pleasing rendition of, e.g an image represented by such information.
In this regard, various schemes are known for performing halftone detection. See, for example, T. Hironori, False Halftone Picture Processing Device, Japanese Publication No. JP 60076857 (1 May 1985); I. Yoshinori, I. Hiroyuki, K. Mitsuru, H. Masayoshi, H. Toshio, U. Yoshiko, Picture Processor, Japanese Publication No. JP 2295358 (Dec. 6, 1990); M. Hiroshi, Method and Device For Examining Mask, Japanese Publication No. JP 8137092 (May 31, 1996); T. Mitsugi, Image Processor, Japanese Publication No. JP 5153393 (Jun. 18, 1993); J.-N. Shiau, B. Farrell, Improved Automatic Image Segmentation, European Patent Application No. 521662 (Jan. 7, 1993); H. Ibaraki, M. Kobayashi, H. Ochi, Halftone Picture Processing Apparatus, European Patent No. 187724 (Sep. 30, 1992); Y. Sakano, Image Area Discriminating Device, European Patent Application NO. 291000 (Nov. 17, 1988); J.-N. Shiau, Automatic Image Segmentation For Color Documents, European Patent Application No. 621725 (Oct. 26, 1994); D. Robinson, Apparatus and Method For Segmenting An Input Image In One of A Plurality of Modes, U.S. Pat. No. 5,339,172 (Aug. 16, 1994); T. Fujisawa, T. Satoh, Digital Image Processing Apparatus For Processing A Variety of Types of Input Image Data, U.S. Pat. No. 5,410,619 (Apr. 25, 1995); R. Kowalski, D. Bloomberg, High Speed Halftone Detection Technique, U.S. Pat. No. 5,193,122 (Mar. 9, 1993); K. Yamada, Image Processing Apparatus For Estimating Halftone Images From Bilevel and Pseudo Halftone Images, U.S. Pat. No. 5,271,095 (Dec. 14, 1993); S. Fox, F. Yeskel, Universal Thresholder/Discriminator, U.S. Pat. No. 4,554,593 (Nov. 19, 1985); H. Ibaraki, M. Kobayashi, H. Ochi, Halftone Picture Processing Apparatus, U.S. Pat. No. 4,722,008 (Jan. 26, 1988); J. Stoffel, Automatic Multimode Continuous Halftone Line Copy Reproduction, U.S. Pat. No. 4,194,221 (Mar. 18, 1980); T. Semasa, Image Processing Apparatus and Method For Multi-Level Image Signal, U.S. Pat. No. 5,361,142 (Nov. 1, 1994); J.-N. Shiau, Automatic Image Segmentation For Color Documents, U.S. Pat. No. 5,341,226 (Aug. 23, 1994); R. Hsieh, Halftone Detection and Delineation, U.S. Pat. No. 4,403,257 (Sep. 6, 1983); J.-N. Shiau, B. Farrell,Automatic Image Segmentation Using Local Area Maximum and Minimum Image Signals, U.S. Pat. No. 5,293,430 (Mar. 8, 1994); and T. Semasa, Image Processing Apparatus and Method For Multi-Level Image Signal, U. S. U.S. Pat. No. 5,291,309 (Mar. 1, 1994).
While there is a substantial volume of art that addresses various issues associated with halftone generation and detection, there has not heretofore been available a fast and efficient technique for effective image classification and for detection of halftone segments and other components of a document. In particular, such techniques as are known do not effectively detect halftone information and classify image regions, especially with regard to efficient algorithms based upon such factors as boundary sets for image information within a single image plane and cross color differences for image information across multiple images planes.
It would be advantageous to provide an improved technique for image classification and halftone detection.
It would also be advantageous to provide a technique that has the to detect halftone components of a document without having predetermined information about the halftone technique used to produce the original image, and moreover, without having detailed information on the specific characteristics of that halftone technique, such as the type of screen used, the threshold array, or the Ipi.
SUMMARY OF THE INVENTION
The invention provides a method and apparatus for image classification and halftone detection.
In a first embodiment of the invention, image classification and halftone detection is performed based on the size of a boundary set, and further based upon image information contained within a single image plane (i.e within one color plane). This embodiment of the invention is based upon the distinctive property of images that halftone areas within the image have a larger boundary set than non-halftone areas within the image.
For example, consider a window of size K×K. In this example, a threshold T1 is adaptively determined and all pixels having a value <T1 are declared to be dark, while all other pixels are declared to be light. This threshold may be set in any of several ways that may include, for example a histogram technique: a histogram of values may be computed in the current window. A right peak area and left peak area are then found in the histogram. If these two areas merge, the threshold is set to the median, otherwise the threshold is set to the end of the larger peak.
As an alternative to the adaptive threshold, another technique, based on a weighted support decision mechanism, can be used to mark the pixels as dark or light.
Given another threshold T2 and a window in the image, the number of vertical class changes and horizontal class changes which occur in the window is counted, where “class change” means a change from a dark pixel to a light pixel or from a light pixel to a dark pixel. The percentage of light pixels in the window is denoted as p, while the percentage of dark pixels is denoted as q. The expected number of vertical and horizontal changes on a K×K window is 4 p q K (K−1).
The type of a current pixel is determined by comparing the actual number of class changes to the probability based estimate. If the ratio of these two numbers is higher than the threshold T2, then the pixel is declared a halftone pixel.
In a second, equally preferred embodiment of the invention, cross color difference correlation is used to detect halftone pixels. This is in contrast to most prior art techniques in which halftone detection and image region classification methods are applied separately to each color component.
In this embodiment, for each pixel having R, G, and B components in an image, there is a surrounding K×K window (K odd). The RGB values of the pixels in this window are denoted R(i), G(i), and B(i), where i=0, . . . , k*k−1; and the RGB averages are denoted aR, aG, and aB. It has been empirically determined that a window size of K=3 or K=5 provides the best results in terms of cost/performance.
The Euclidean norms of the R( ), G( ), B( ) vectors are denoted IRI, IGI, IBI, and the following sums are computed:
xRG=&Sgr;((R(i)−R)(G(i)−G)) xaRG=&Sgr;((R(i)−
Karidi Ron J.
Nordstrom Niklas
Electronics For Imaging, Inc.
Glenn Michael A.
Lee Thomas D.
LandOfFree
Method and apparatus for image classification and halftone... 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 and apparatus for image classification and halftone..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for image classification and halftone... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2576520