Correction of distortions in form processing

Details Correction of distortions in form processing Correction of distortions in form processing

2000-03-08

2003-12-02

Boudreau, Leo (Department: 2621)

Image analysis

Image transformation or preprocessing

Selecting a portion of an image

C382S276000, C358S452000

Reexamination Certificate

active

06658166

ABSTRACT:

FIELD OF THE INVENTION
The present invention relates generally to document image processing, and specifically to correction of distortions in document images.
BACKGROUND OF THE INVENTION
In many document imaging systems, large numbers of forms are scanned into a computer, which then processes the resultant document images to extract pertinent information. Typically the forms comprise preprinted templates, containing predefined fields that have been filled in by hand or with machine-printed characters. To extract the information that has been filled in, the computer must first identify the fields of the template and then decipher the characters appearing in the fields. Various methods of image analysis and optical character recognition (OCR) are known in the art for these purposes.
In order to identify the fields of the template and assign the characters to the correct fields, a common technique is for the computer to register each document image with a reference image of the template. Once the template is registered, it can be dropped from the document image, leaving only the handwritten characters in their appropriate locations on the page. Such registration may be difficult, however, because of distortions introduced in scanning the original form, including skew, rotation, warp and other scale changes.
One method for dealing with these distortions is described in U.S. Pat. No. 5,182,656, whose disclosure is incorporated herein by reference. The original image is partitioned into a number of relatively small overlapping segments. Each of the segments is then shifted in order to bring it into alignment with an appropriate, corresponding segment of the reference template image. More complex transformations, such as rotations or scale changes, are not performed on these segments. The transformation of the entire image is thus represented as a combination of the shifts of the small segments, which can approximate rotations and scale changes if the segments are made small enough.
FIG. 1
is a schematic illustration showing a detail of an image of a filled-in form document
20
, useful in understanding the method of U.S. Pat. No. 5,182,656 and of similar methods known in the art. Here a name, comprising characters
24
, has been filled into a box
22
provided by a template on document
20
. The box is slightly rotated relative to its proper, horizontal position on the reference template. In order to correct this rotation and for other distortions in the scanned image, the image of the document is divided into segments
26
,
28
,
30
,
32
, etc. The image is analyzed to determine the appropriate shift transformation to be applied to each of the segments, as specified in the patent.
FIG. 2
is a schematic illustration of segments
26
.
28
,
30
and
32
in their respective, transformed positions. To compensate for the rotation, a different shift is applied to each of the segments. The relative shifts are exaggerated in the figure for clarity of illustration. Furthermore, an overlap has been introduced between the transformed segments, such as may result from a scale distortion in the scanned image, for example. In these areas, the above-mentioned patent suggests performing an OR operation for each pixel in order to avoid having one segment overwrite another.
FIG. 3
schematically illustrates the results of the shifts shown in FIG.
2
. Characters
24
are broken into respective upper portions
34
and lower portions
36
, because the characters happened to cross the boundaries between segments
26
and
30
and between segments
28
and
32
. In consequence, even though box
22
may in the end be successfully registered with the corresponding box in the reference template image, the characters in the box are difficult or impossible for the computer to decipher. This document will probably have to be passed to a human operator for data entry, adding substantially to the cost of processing the document. What is worse, the computer may misinterpret the distorted characters, leading to an error in the data extracted from the form.
U.S. Pat. No. 5,793,887, whose disclosure is incorporated herein by reference, describes another method for alignment of images for template elimination. In this case, a filled-in document image and a reference template image are divided into vertical bands. The bands are shifted relative to one another in order to correlate the lines in the document image with the lines in the corresponding bands of the reference template image. If necessary, the procedure is then repeated using horizontal bands. Thus, this method can also lead to break-up of characters, as illustrated in FIG.
3
.
SUMMARY OF THE INVENTION
It is an object of the present invention to provide improved methods and apparatus for processing images, and particularly for processing images of filled-in form documents.
It is a further object of some aspects of the present invention to provide methods and apparatus for document image processing that improve the readability of characters in such images in the presence of image distortion and rotation. It is a particular object of these aspects of the present invention to alleviate problems of readability that may arise due to misalignment between segments of such images that are shifted for the purpose of template registration.
In preferred embodiments of the present invention, an input document image, containing characters filled into a form template, is processed in order to register the template in the image with a reference template image. Any suitable method known in the art may be used for this purpose. Regions of interest, defined as regions containing filled-in characters, are identified in the processed image. Each of these regions is preferably checked in order to determine whether the readability of the characters in the region has been adversely affected by transformations applied in processing the image to register it with the reference template. Typically, although not exclusively, such adverse effects result when different parts of the region fall into different segments of the image to which different transformations are applied. The contents of each of the affected regions (or of all of the regions) are then erased from the processed image and are replaced by the contents of the corresponding region of the input image. In this manner, all of the characters in the input image are placed in the correct locations relative to the reference template, substantially without adverse effect on the readability of the characters.
Although preferred embodiments are described herein with reference to document form images, it will be appreciated that the principles of the present invention may likewise be applied in other contexts, as well. It frequently occurs in image processing that a digital transformation applied to an image is not exact, in the sense that the resulting offset between two neighboring pixels is different from the theoretical offset. This inexactness may be due to local segment transformations, as described above, or to other errors, such as rounding or decimation. If there are areas of particular interest in the image, such as a text block or other significant image features, the method of the present invention may be used advantageously to enhance the readability and/or clarity of details in these areas.
There is therefore provided, in accordance with a preferred embodiment of the present invention, a method for processing an input image, including:
applying one or more transformations to the input image, whereby different shifts are applied to different pixels in the input image, so as to generate an output image;
selecting in the output image a region containing content of interest;
locating in the input image the region corresponding to the selected region in the output image; and
substituting the content of the located region in the input image for the content of the selected region in the output image.
Preferably, applying the one or more transformations includes dividing the input image into segments, and determining

Affiliated with

Also associated with

Rating

Correction of distortions in form processing does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Correction of distortions in form processing, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Correction of distortions in form processing will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3141407