Marks Based OCR =============== Relatively complete document image analysis system that largely ignores image information during the classification stage. Instead relies on contextual and language cues. Developed by Scott Leishman for research purposes while attending U of T. 1. INSTALLING ------------- 1.1 - Requirements ------------------ * Matlab 7 or higher (earlier versions won't work due to use of imview and find(...'first') etc.) 1.2 - Notes ----------- * Copy all files (and subdirs) in this directory to some location on the users computer (note this location -- we'll denote it INS_PATH). * Add the following lines to your ~/matlab/startup.m (or type it from the Matlab command prompt): global MOCR_PATH; MOCR_PATH = 'INS_PATH'; addpath(MOCR_PATH); marks_ocr_init; (where INS_PATH is replaced by the actual path to where the files and subdirs were installed) 2. Configuration ---------------- Typically, modifiable parameters are found at the top of key files before being passed on to other functions as neccessary. Descriptions of each parameter are located within such a file and sensible defaults are already defined. * for clustering and connected components, see cluster/cluster_comps.m * for line finding, see util/get_lines.m * ... 3. Running ---------- * There are 3 main tasks: creating training data, clustering components of file images and constructing a language model from them, and performing ocr on lines of training data, given a constructed language model. * To create new training cases, the following commands are useful: * ...
scttl/marks_based_ocr
Matlab based document image analysis and classification system, that makes heavy use of contextual and language cues to decode image glyphs
MATLABMIT