homedemos & tools › pixlabeler

DEMOS & TOOLS:

PixLabeler
pixel-level labeling of document image content

 

 

WHAT: a program that enables pixel-level labeling of document image content; foreground markings in user-selected regions (by enclosure, by brushing, or point-and-click) are colored according to label type

WHY: to intuitively, easily, and economically prepare large amounts of "groundtruth" data for training and evaluating document recognition programs (such as those designed to separate handwritten markings, machine print, machine graphics, speckle noise, etc.)

HOW: combines customized user interface design with background processes that possess knowledge of structure in document images; applies line art recognition and connected component analysis to form groups of pixels that can be selected by point-and-click; also supports bootstrapping by existing automatic labeling algorithms

TRY IT OUT