8 4 7 7 8 0 0 1 0 0

The examples in Figure 16 and Figure 17 demonstrate how system errors have a cascading effect, resulting in multiple errors being introduced into a single hypothesis string. The alignment examples shown are, by design, easy to understand and are easily derived. In practice, multiple errors frequently occur in a single hypothesis string resulting in many different possible alignments. The Scoring Package analyzes each candidate alignment and chooses the one that assesses the least amount of penalty. The Scoring Package does this in a consistent and logical way so that, when given the same hypothesis string and reference string, the Scoring Package will always generate the same alignment. As multiple errors are introduced into the hypothesis string, it becomes increasingly more difficult for the Scoring Package to unambiguously distinguish insertion errors from substitutions errors. This distinction often requires human inspection which would compromise the degree to which the Scoring Package is automated. Therefore, the Scoring Package does not distinguish substitution errors from insertion errors and lumps them together into a single category called false positives. 4 5 6 7 8 4 7 7 8 0 0 1 0 0 Images 0 1 0 0 0 Alignment of Substitutions (1= Sub) 22 result of segmenting the isolated field into separate images, which are assumed to be one character per image. Notice the '4' has been incorrectly separated into two pieces resulting in two isolated images with two strokes forming a right angle in the left image and a vertical stroke in the right image. This is an example of a segmentation failure, the splitting of a character into multiple images. The third line in the figure lists what was printed in the field. The fourth line lists the hypothesis string corresponding to the assigned classifications generated by the recognition system. Figure 16: Scoring Package alignment of an insertion error caused by a segmentation failure and resulting in a substitution error. Due to the segmentation failure, the character classifier in the recognition system is presented the two pieces of the '4' rather than one complete character. The result can be seen in the hypothesis string where the first piece of the '4' is classified as a '6' and the second piece of the '4' is classified as a '1'. The fifth line in the figure marks the insertion error identified by the Scoring Package with a '1' representing the inserted classification of a …