The design and development of the classifier have been presented. The concept exploration dataset was used to identify potentially good indicators of image defects from an OCR point of view. A subset of these indicators were selected to be included in the classifier because of their simplicity. Tentative metrics were proposed to measure these indicators and the training dataset was used to determine the actual form of these metrics. Finally, from the metrics developed, a set of heuristic rules was put together to implement the classifier's logic.