Improved Photo-Document Segmentation

Laserfiche Computer Science, 2006-07

Liaison(s): Karl Chan ’87, Ed Heaney ’04, Kurt Rapelje
Advisor(s): Zachary Dodds
Students(s): Adam Field, Stephen Smith, Benjamin Tribelhorn (PM), Aaron Wolin

This project seeks to replace a document scanner with a hand-held digital camera as a front-end to Laserfiche’s document-management solutions. Conditions such as shadows, indistinct backgrounds, multiple documents, and occlusions make it difficult to determine the document’s location within the camera’s image: this is the segmentation problem. The team has integrated pixel-level, edge-level, and structural-level image processing routines within a probabilistic framework to find appropriate document segmentation in a variety of business cases.