System and method for automatic document segmentation

• US 5,073,953 A
• Filed: 09/01/1989
• Issued: 12/17/1991
• Est. Priority Date: 09/12/1988
• Status: Expired due to Term

• Current Assignee: Oce-Nederland B.V. (Canon Inc.)
• Original Assignee: Oce-Nederland B.V. (Canon Inc.)
• Inventors: Westdijk, Jacob A.
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Stellrecht, Barry
• Application Number:
  US07/403,195
• Time in Patent Office: 837 Days
• Field of Search: 382/9, 382/51, 382/48, 358/462, 358/457, 358/453
• US Class Current: 382/176
• CPC Class Codes: H04N 1/40062 Discrimination between different image types, e.g. two-tone, continuous tone

Abstract

In a system and method for automatically segmenting a scanned document to separate areas containing different types of information such as black/white text or graphics, continuous tone pictures, and half-tone pictures, the document is divided into a number of subimages and the individual subimages are classified in an initial labeling phase. The initial label matrix thus obtained is relaxed in a subsequent step so that a pattern of uniformly labeled segments corresponding to the areas in the document containing different types of information is the result. In order to speed-up the system and/or improve the robustness thereof, the number of initial labels used in the initial labeling step is selected larger than the number of types of information to be distinguished. The number of initial labels is then reduced in the relaxation step on the basis of a series of context rules.

Claims

1. A method for automatically segmenting a scanned document in an electronic document processing device to separate document areas containing different types of image information comprising the steps of:

(a) scanning the document to generate a scanned image thereof;

(b) subdividing the scanned image into a matrix of subimages;

(c) analyzing the information contained in each subimage and assigning to each subimage an initial label selected from a first set of labels to obtain an initial label matrix; and

(d) relating the initial label matrix by changing the labels of individual matrix subimages into relaxed labels which are selected from a second set of labels, pursuant to a plurality of context rules, which is smaller in number than said first set to obtain a pattern of uniformly labeled segments representing those document areas containing different types of information, said context rules comprising the following;

(i) if in a predetermined array of subimages in the initial label matrix at least n subimages have the label BW and this array does not contain labels from a group G, then change all labels in this array to BW", wherein n is a predetermined number, BW is a predetermined initial label and G is a predetermined subset of the first set of labels;

(ii) "if in a predetermined array of subimages in the initial label matrix at least m subimages have the label U, then change all labels in this array to U, " wherein m is a predetermined number and U is a predetermined initial label; and

(iii) (1) "where the label U forms intersecting vertical and horizontal runs, fill the whole rectangle spanned by these runs with the label U,

(2) check all combination of horizontal and vertical runs to maximize the area to be filled with the label U, and

(3) if the height of the maximized area is smaller than h_min elements or the width is smaller than w_min elements, then change all labels within this areas to BW, " wherein U and BW are predetermined initial labels and h_min and w_min are predetermined numbers.

2. The method as described in claim 1 wherein the context rules further comprise the following:

"if the rectangle, mentioned in (iii) (1), contains a number of U labels greater than a predetermined number Umin, fill the whole rectangle with label U," and

"if the shape of the rectangle, mentioned in (iii)(1), fulfills the conditions;

width/height>

min. and width/height<

max., fill the whole rectangle with label U;

"

wherein min. and max. are predetermined numbers.

3. The method as described in claim 2 wherein the predetermined arrays to which the context rules (i) and (ii) are applied have a size of three by three subimages.

4. The method as described in claim 1 wherein the relaxation step further comprise a plurality of steps for step-wise modifying the initial label matrix by applying each context rule at least one time pursuant to a predetermined sequence.

5. The method as described in claim 1 wherein the context rule with the structure as defined in (i) is applied prior to the context rule with the structure as defined in (ii) and the context rule with the structure as defined in (ii) is applied prior to the context rule with the structure as defined in (iii).

6. The method as described in claim 5, wherein the context rules further comprise a LOCAL rule prescribing that:

"if a given matrix element is surrounded by upper, lower, right and left immediate neighbors which all have the same label X, the label of the given matrix subimage is also changed to X" and wherein the LOCAL rule is applied immediately before each of the context rules with the structure as defined in (i), (ii), or (iii).