• Sub Navigation

Ascent Capture:
Product Support Overview
License Activation
Release Scripts
Temporary AC Activation Code

OTHER TOOLS:
Knowledgebase
Support Request

PRODUCT INFO:
Product Overview

 

Ascent Capture 6.x Application Note

AC 7.5 is the latest release of the Ascent Server product.

---

Kform vs. OCR Registration for Page Registration and Page Recognition

Purpose: This document explains Kform vs. OCR registration for Page Registration and Page Recognition and the value of each method.

Page-Level Matching Processes in Capture 6.x

Page-Level Registration

By default, new batch classes have a "page matching" algorithm enabled for all pages with index zones. This algorithm compares the sample page to the scanned image using an image controls module called Kform. This process can verify whether a scanned image "looks like" the sample image, and also automatically adjust zones based on any horizontal or vertical shifting (also called registration) introduced by the scanner.

This process is automatic and requires no setup. In general, it helps to provide better registration of zones (yielding better OCR results), and a good verification that the scanned images are of the expected format and are not rotated.

However, there are cases where you will want to turn this feature off. To disable the feature, you must edit the properties of the form type and set the page level registration confidence slider to 0. Note that 2.x batch classes have this property turned off by default (since it is likely that such batch classes already meet one of the criteria listed below). Here are the situations when you would want to turn this feature off:

• The scanned images are stretched and/or distorted. This will typically cause the page matching algorithm to fail causing various pages to be rejected in forms processing. In this case, you cannot use the page matching algorithm. It is suggested that you create an OCR registration zone to properly offset zones. See the related document on zonal processing for more information on this.
• The scanned images vary with respect to the sample image. This might occur when the customer has certain parts of the image that are standard (like a header from which they extract OCR zones), but has other parts of the image that vary from one page to the other. In this case, no matching can be performed. Again, OCR registration on the static portion of the page is recommended.
• The scanned images vary in DPI or have a different DPI than the sample images. The page matching algorithm cannot deal with differing DPI ratings. A work-around for this is to create multiple form types with identical information, but with sample pages scanned at the various different DPI's expected. If there is not a potential form type that matches the scanned DPI, then the Scan module provides a warning that the pages will be rejected in forms processing. If unknown sets of DPI combinations are expected, and the above work-around is not sufficient, then disable page level registration and utilize OCR registration instead.
• The scanned images match the background of the sample form, but are much "busier." An example of this might be a sparse blank form with few lines on it. Actual scanned images have text filled in. In such cases, the page matching algorithm can become confused since the filled-in text is mistaken for static "features." Better results can sometimes be realized by utilizing sample images with actual data filled in. Also, the filled in information can be "cleaned" by using aggressive despeckling. If these techniques don't work, then page-level registration must be disabled.
• Performance is critical and an OCR registration zone can be used reliably. Page-level registration takes about 1 second per page whereas OCR registration takes about 1/3 of a second. Given that average documents only take about 3-6 seconds to process, this added time can be significant. Turning off page-level registration eliminates the 1 second performance hit, but may not allow for sufficiently accurate zonal results (assuming scanned pages are shifted). It is suggested that OCR registration be used in this case.
• The user gets random page matching errors in Forms Processing that cannot be explained by any of the above issues. In this case, try trimming the page-level confidence to lower values. Another strategy might be to try rescanning the sample images. It has been noted that scanning the sample images on a scanner identical to the ones used for production scanning improves the reliability of this feature. If these techniques don't work, then page registration must be turned off and OCR registration should be used instead. In these cases, the problem along with the sample image and offending scanned images should be sent to Kofax since the algorithm should work with all images in theory (given future enhancements).

Page-Level Form Identification

An option for performing form identification (and potentially separation as well) is Kofax Page Identification. This name is slightly confusing because the process actually consists of two steps:

1. The scanned image is matched to the first sample page of each potential form. This algorithm is identical to that used for page registration (described above).
2. Any scanned images that "pass" the page-level phase are searched for form id zones. Any forms with zones that cannot be recognized are disqualified. Forms with zones that match are considered the "winning" forms. Typically, there is only one "winner", but if multiple forms qualify, then the one with the highest page-level confidence is taken.

For more detail on this process, see the related document on form identification.

In most cases, Form ID zones are not necessary provided the sample forms look different from each other. If some of the forms look the same, then Form ID zones should be added (usually OCR zones) to differentiate those forms. In some cases, Form ID zones are required on all the forms.

In such cases, it is best to turn off the page-level phase completely as doing so improves performance and eliminates false errors. To do this, set the page-level confidence in the "Custom Separation and Form Identification" dialog box to 0. Setting this confidence to 0 and creating Form ID zones on all forms is a good idea in the following situations:

• All the sample forms look similar to each other.
• The scanned images are stretched and/or distorted.
• The scanned images vary with respect to the sample image.
• The scanned images vary in DPI or have a different DPI than the sample images. The same work-around as suggested for registration may also work in this scenario.
• The scanned images match the background of the sample form, but are much "busier."

The elimination of the 1 second per page overhead for page-level recognition is deemed sufficiently important to warrant the added setup time required to create reliable Form ID zones.