Select the method you want to use for determining the confidence level assigned to the results generated by the OMR engine.

All of these methods make use of the ratio of black to white pixels within the zone. When setting up the zone, keep the following in mind:

• If you draw a large zone box around an OMR item, the percentage of black pixels is relatively small. Consequently, if a user makes a small check mark, it may not significantly change the percentage of black pixels. If this small change does not put the percentage over the threshold, then the OMR engine detects it as unmarked. Therefore, the smaller the zone box around the OMR item, the easier it is to detect a mark. This is the confidence method used in all previous versions of Kofax Capture.

• There are always slight differences in registration (paper position) as documents are fed through a scanner. If your zone box is very close in size to the OMR item, then the paper may shift so that part (or all) of the item is no longer in the zone, leading to detection errors. Therefore, the smaller the zone box around the OMR item, the greater the chance of registration problems.

• For best results, initially make the zone boxes as small as you can. If registration errors turn out to be a problem, gradually increase the size of the box until the errors are minimized. Adding additional registration zones may also help to increase accuracy.

• Make sure that zone boxes around similar OMR items are all the same size, to ensure consistent results.

Pixel percentage filled

This is the default setting. The confidence is calculated as the percentage of black pixels in the OMR zone. For example, a confidence of 32 percent means that 32 percent of the pixels in the OMR zone are black. The higher the number, the more likely the box has been checked.

This graph shows the relationship between pixel percentage and confidence for the pixel percentage filled method of determining confidence. In this case, the threshold is set to 50 percent, but it could be set anywhere between 0 percent and 100 percent.

Probability that zone is marked

The confidence is calculated based on the probability that the zone has been marked. Typical confidences for marked zones fall in the 80-90 percent range and unmarked zones in the 0-20 percent range. This method is well-suited for zones used in groups when Pick best is checked in the Index Group Zone Properties window.

This graph shows the relationship between pixel percentage and confidence for the probability that zone is marked method of determining confidence. In this case, the threshold is set to 50 percent, but it could be set anywhere between 0 percent and 100 percent. Moving the threshold displaces the center of the curve left or right, but the basic shape remains unchanged.

100% if marked, 0% if not

This is similar to the preceding option. The confidence is set to either 100 percent or 0 percent (no intermediate values) depending on whether the zone is marked or not.

This graph shows the relationship between pixel percentage and confidence for the 100 percent if marked, 0 percent if not method of determining confidence. In this case, the threshold is set to 50 percent, but it could be set anywhere between 0 percent and 100 percent.

Note The threshold setting and the confidence are not related in this case.

Difference from threshold

The confidence is calculated on how far the pixel percentage is from the threshold point. Consequently, a completely blank zone and a completely filled zone would both have a confidence of 100 percent. On the other hand, if the threshold is set to 30 percent and the number of black pixels is 29 percent or 31 percent, the confidence level is very low, since a small error in scanning could flip the results. This method indicates how certain the OMR engine is of its results, not the pixel percentage.

This setting is a good choice for yes/no type check boxes that are not part of a group. In general, using this setting in a group does not produce useful results and is not recommended. In groups, you may need the recognition engine to pick the best result. Since this method can yield 100 percent confidence for both a blank and a checked item, your pick best results cannot be predicted.

This graph shows the relationship between pixel percentage and confidence for the difference from threshold method of determining confidence. In this case, the threshold is set to 50 percent, but it could be set anywhere between 0 percent and 100 percent. Moving the threshold displaces the center of the curve left or right, but the basic shape remains unchanged.