The Headstone Photo Processing System is a desktop application that automates the tedious work of matching photographs of headstones to their official cemetery records. The project grew directly out of a problem facing its sponsor, Dr. Amy Giroux, who previously processed thousands of cemetery photos by hand—cropping and straightening each image, reading the engraved text, and matching every headstone to the correct entry in a records spreadsheet. This system handles that pipeline automatically through three connected stages: a machine learning model (Mask R-CNN) that detects each headstone in a photo and crops and rotates it to a clean, upright image; an optical character recognition module (built on AttentionOCR) capable of reading even the curved and weathered text common on older headstones; and a fuzzy-matching search that links the recognized text to the right cemetery record even when erosion makes a few characters unreadable. Anything the system can’t confidently handle is routed to a “failure queue” for a human to review, ensuring accuracy is never sacrificed for automation. Tested across several national cemeteries—including Alexandria, St. Augustine, Florida National, and Fort Meade—the tool dramatically reduces the manual workload of preserving and digitizing veterans’ grave records.
