UC Santa Cruz

Printed text accessibility is an issue that impacts mobility, scholastic achievement, and thus career growth. Our research addresses how to make printed text accessible to a blind person.

We start this dissertation with a formal definition of the range of positions and orientations from where an OCR readable document image can be acquired from. Initially, we conducted a naive user studies with a wizard-of-oz system that relied on fiducial markers printed on a document to facilitate visual odometry in order to study if blindfolded people could capture compliant document images with the help of our system. We then moved on to blind participants with an experiment meant to simultaneously assess the range of problems with document image acquisition and test different interaction modalities statistical effect on the time to capture an image.

Finally we devised a computer vision algorithm, without fiducials, capable of verifying document image compliance and providing instructions for acquisition. We tested this algorithm with participants in a redesigned counterbalanced repeated measures experiment. Our analysis revealed that guidance significantly reduces the amount of time necessary to capture a compliant image. Our participants expressed positive comments on the system, and generally felt that their proficiency at taking OCR-readable images had increased by interacting with the system.