Computer Vision and Federal Census Data

Our final presentation will showcase our efforts to duplicate large demographic data development for other historical communities. In addition to writing programs to identify the same individuals from different census years over time, we are also experimenting with image processing and computer vision technology, hoping to partially automate the process of data collection from scanned images of handwritten census forms. Optical character recognition software, which works well for printed text, is not promising for the handwritten census forms. However, our work to categorize images of individual data items and group similar images together has the potential to simplify and automate the creation of large demographic datasets like those of the Digital Harrisburg project.