Manual Correction of Misclassified ImageSegments

摘要

Since book scanning is often used for low-volume titles, controllingscan cost is critical. Automatic image segmentation promises high quality andlow cost scans with a minimal need for user intervention. However, since nosegmentation algorithm is perfect, strategies based on business policies arerequired to manage trade-offs between cost and quality. We describe an adaptiveapproach that allows the user to control, based on policy, the amount ofintervention necessary to verify the segmentation’s correctness and to rapidlyadjust misclassified segments.