Method and apparatus for transforming an image for classification or pattern recognition

• US 6,246,793 B1
• Filed: 11/03/1997
• Issued: 06/12/2001
• Est. Priority Date: 02/03/1993
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Rudolph, Volker, Rindtorff, Klaus
• Primary Examiner(s): Bella, Matthew C.
• Application Number:
  US08/962,786
• Time in Patent Office: 1,317 Days
• Field of Search: 382/168, 382/174, 382/182, 382/185, 382/187, 382/190, 382/203, 382/289, 382/199
• US Class Current: 382/174
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/15 Cutting or merging image elements, e.g. region growing, watershed or clustering-based techniques
• CPC Class Codes: G06V 30/18095 Summing image-intensity values; Projection and histogram analysis
• CPC Class Codes: G06V 30/2264 using word shape
• CPC Class Codes: G06V 40/30 Writer recognition; Reading and verifying signatures

Abstract

The invention relates to a method and apparatus for transforming an image for classification or pattern recognition. At least two distinct projections are carried out in order to define boundaries and shadows of the transformed image. The transformed image is the basis for further analysis such as the calculation of parameter values to be verified.

Claims

1. A method of transforming an image for pattern recognition comprising the steps of:

a) defining a first boundary of said image by a first projection;

b) defining a secondary boundary of said image by a second projection, said second projection being distinct from said first projection and defining one or more further boundaries of said image by one or more further projections, one of said further projections resulting in a minimal shadow, said one of said further projections having a reference line substantially parallel to a slant of a character of said image;

c) defining a transformed image by said boundaries of said image and resulting shadow or shadows of said projections;

d) repeating at least one of said projections while ignoring one of said boundaries of said image; and

e) repeating step d) iteratively until a termination condition is fulfilled.

2. A method of transforming an image including the steps of claim 1, further comprising the steps of:

f) searching for a portion of the border of at least one of said shadows, which is not limited by one of said boundaries;

g) deleting at least a portion of said shadow by a projection having a reference line substantially orthogonal to said portion of the border; and

h) carrying out steps f) and g) iteratively until a termination condition is fulfilled.

3. A method of transforming an image including the steps of claim 2, further comprising the steps of:

i) assigning a first identifier to that portion of said shadow which is deleted due to the projection carried out in step of deleting at least a portion of the shadow; and

j) assigning an ith identifier to that portion of said shadow which is deleted due to the ith projection carried out in the ith iteration of deleting the portion of the shadow.

4. A method of transforming an image including the steps of claim 1 and further comprising the steps of:

f) identifying a region of said transformed image being completely circumscribed by a number of said boundaries;

g) opening said circumscribed region so that a portion of a shadow being comprised in said region is no longer limited by one of said boundaries;

h) deleting at least a portion of said shadow by a projection having a reference line substantially orthogonal to said portion of the border; and

i) carrying out the steps f), g) and h) iteratively until a termination condition is fulfilled.

5. A method of transforming an image including the steps of claim 1 wherein said termination condition includes one chosen from the following conditions of a maximum number of iterations, a minimum size of a remaining shadow, a minimum change in the size of the remaining shadow from one iteration to the next and a maximum computing time and said termination condition is fulfilled, if no shadow is left.

6. A method of text segmentation comprising the steps of:

a) transforming a digitized image of said text according to the method of claim 1;

b) detecting gaps in said transformed image; and

c) segmenting said text into text segments and gaps, such as words or individual characters, whereby said gaps deliminate each segment.

7. The method of text segmentation including the steps of claim 6, wherein the text includes a slant and one of said projections has a reference line being substantially parallel to the slant of the text to be segmented.

8. Apparatus of claim 1, further comprising:

means for searching for a portion of the border of at least part of said shadow which is not limited by one of said boundaries;

means for deleting at least a portion of said shadow by a projection having a reference line substantially orthogonal to said portion of the border.

9. An apparatus for transforming an image for pattern recognition, comprising:

a) means for defining a first boundary of said image by a first projection;

b) means for defining a second boundary of said image by a second projection, said second projection being distinct from said first projection;

c) repeating at least one of said projections while ignoring one of said boundaries of said image;

d) repeating step c) iteratively until a termination condition is fulfilled;

e) means for defining a shadow defined by said first and second boundaries and for using said shadow in pattern recognition, whereby the transformed image is defined by said boundaries and one resulting shadow of said projections; and

f) means for defining a further boundary of said image by a further projection, said further projection resulting in a minimal shadow and having a reference line substantially parallel to a slant of a character of said image.