Image analysis – Image segmentation – Distinguishing text from other regions
Reexamination Certificate
1999-12-08
2004-05-04
Wu, Jingge (Department: 2623)
Image analysis
Image segmentation
Distinguishing text from other regions
C382S239000, C358S464000
Reexamination Certificate
active
06731800
ABSTRACT:
The invention relates to a method for compressing scanned, colored and gray-scale documents, the digital image of the scanned document being divided into three image planes. These image planes are a foreground image, a background image and a binary mask image. The mask image describe which areas of the document belong to the foreground and which to the background. Components of the foreground are text and graphic elements. The color and intensity of these foreground areas is described by the foreground image. The background areas include the text background as well as the pictures contained in the document. The color or brightness information of the background as well as the images contained in the document are contained in the background image. Each of the three images is coded separately with a suitable image coding method. For the decoding, the document is assembled once again from the foreground and background images. In this connection, the binary mask image describes the areas, in which the reconstructed document is to be generated from the foreground image or from the background image.
In the case of a suitable division of the document into the three image levels described, clearly better compression results can be obtained with this representation than with image coding methods which code the document as a whole. The invention describes a new method for determining the binary mask image as well as a method for the efficient division of the original document into the foreground and background images. No special assumptions are made regarding the nature or construction of the document.
STATE OF THE ART
Documents, which are scanned with high resolutions, require much memory for the resulting digital image data. For example, a 300 dpi scan of a colored A4 page takes up approximately 25,000,000 bytes and a colored 600 dpi scan takes up 100,000,000 bytes. Documents with amounts of data of this order of magnitude can be archived uncompressed only in small amounts. Transfer over networks with low transfer rates is practically impossible.
Lossless compression methods, such as the lossless mode of the JPEG standard (JPEG-LS) or Lempel-Ziv Welch (LZW) only make very small compression factors possible. Higher compression factors are possible only through the use of loss-affected compression methods. The DCT-based “JPEG method” of the Joint Pictures Expert Group is considered to be the standard method. However, neither the JPEG method nor the newer, better wavelet-based compression method can be used for the high-grade compression of scanned documents. These strictly image compression methods presume the statistics of typical image signals, which are characterized by a high local correlation. Since these assumptions do not apply to scanned documents, the text portion of the documents is greatly changed at high compression factors, so that it becomes impossible to read the text. At the present time, documents for archiving are usually scanned in the binary mode and compressed with the CCITT fax compression standard “Fax Group 3” or “Fax Group 4”. In general, the readability is retained by these strictly binary compression methods. However, the brightness and color information of the image portions is lost entirely.
The Mixed Raster Content standard (MRC) (ITU recommendation T.44), which is presently in the planning stage, is a new attempt to avoid these problems. According to this standard, it is possible to divide a document into regions of different local resolution, which can be coded in different ways. One mode of the MRC standard is a multi-layer coding mode, which provides for a division of the document into three previously described planes. However, in the MRC standard, exclusively the decoding process is fixed unambiguously. The method of the division of the documents into three image planes during the coding is not specified. A method, which uses this multi-layer coding mode, is described in U.S. Pat. No. 5,779,092. However, the method specifies conditions, which are not applicable in many of the documents, which are to be processed, namely that the shape of the images is assumed to be rectangular. No provisions are made for detecting text with background images or for the presence of text within images. Furthermore, the document background basically must be white or bright.
It is an object of the invention to make possible the compression of scanned documents without being restricted by the nature of the original copies, such as a bright background, rectangular illustrations and precise separation of image and text components. Moreover, the expenditure of computing time and the use of memory shall be reduced clearly. Furthermore, different classes of documents and images shall be compressed according to a uniform method by way of a few control parameters.
SUMMARY OF THE INVENTION
A new method for compressing scanned documents is introduced. Starting out from the representation of the document in three planes—foreground image, background image and mask image—the basic course of the compression method is presented in the following:
To begin with, a locally variable threshold value image is generated from the defined reduced original document with an adaptive threshold method, and brought back once again to the size of the original document. The original image is subsequently quantized with this threshold value image in order to produce a bitonal quantization image. By means of this quantization image and the original image, a text detection (segmenting), which divides the document into foreground and background regions, is subsequently carried out. In so doing, regions such as text and graphic elements are assigned to the foreground image, the text background and the images of the background image. The result of this segmenting is filed in the binary mask image. The binary mask image has the same size and resolution as the original image. Subsequently, a foreground image, which describes the color of the foreground regions, is produced from the original image and the mask image. Compared to the original image, this foreground image has a reduced resolution. After that, the background image is produced once again with a reduced resolution from the complement of the bitonal mask image and the original image. Subsequently, the three images are coded in each case with a suitable image coder.
DESCRIPTION OF THE INVENTION
The quantization or binarizing of the original document takes place in two steps. To begin with, a locally variable threshold value is determined with an adaptive method. The comparison of the gray value representation of the original document with this threshold value supplies the binary quantizing image.
If the original document exists as a colored document, it is converted, to begin with, into a gray scale image and two color difference component images. The gray scale image is used for determining the locally variable threshold value image. The gray scale image, which normally exists with a resolution of 150 to 600 dpi, is reduced to a suitable size. For this purpose, a low-pass filtering with subsequent sub-sampling is carried out. Local distortions in the original scan, such as noise or dither and raster effects are decreased by the reduction.
A local dynamics analysis is carried out next. For this purpose, the reduced gray scale image is subjected to a minimum filtration and a maximum filtration. The difference between the maximum image and the minimum image produces a dynamics image, which supplies indications regarding regions with strong edges, such as text regions.
In the next step, the dynamics image is compared with a minimum dynamic, which can be specified externally and controls the sensitivity of the quantizing and of the text detection. In regions, the dynamics of which exceed this minimum value, a quantizing threshold value is determined from half the sum of the minimum image and the maximum image. In regions, the dynamics of which are too low, the quantizing threshold value initially is set equal to zero.
Since the threshold values, so determ
Barthel Kai Uwe
McPartlin Simon
Thierschmann Michael
Algo Vision Lura Tech GmbH
Jordan and Hamburg LLP
Wu Jingge
LandOfFree
Method for compressing scanned, colored and gray-scaled... 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 compressing scanned, colored and gray-scaled..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for compressing scanned, colored and gray-scaled... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3263936