Toolkit to locate the baselines in handwritten text pages. Use it at your own risk.

• A toolkit to find the text baselines in pages.

• BaseLinPages is written in C++.

• BaseLinPages uses OpenV and Bash shell

• Some features: * Flexible. * Easy and fast to set up. Roughly, around 15 minutes to anotate a page, 3 minutes to train an Extremely Randomized Trees Classifier and 3 seconds per page to get the baselines * Mainly for academic use.

• trainForestNPages.sh corpus.lst CorpusDir NumberOfPages ConfigFile.cnf
  • Change NumberOfPages to the apropiate value (#integer). It takes the NumberOfPages first pages from the corpus.lst

• test.sh corpus.lst CorpusDir PathToErtClassifier ConfigFile
  • Change NumberOfPages to the apropiate value (#integer)

• measures.sh corpus.lst HypotesisDir GroundTruthDir
  • This tool is based on the java program used in ICDAR 2017 Competition on Baseline Detection (cBAD)
  • In this paper the method used is explained.

#local minima

• MINIMA_WINDOW=15
• MIN_POINTS_CONTOUR=35 #by default

#vars for the labelPointsFromBaselines

• CLASS_XML_DOWN_BOUND=8
• CLASS_XML_UPPER_BOUND=10

#vars for the ERT classifier

• MINIM_PROB_TO_TRUE=0
• K=0 #by default

#vars for the mdbscan

• DBSCAN_MIN_DENSITY=2 # by default
• DBSCAN_EPSILON_DIST=400 #0 for automaticaly estimated
• DBSCAN_MIN_POINTS_PER_LINE=2
• DBSCAN_MAX_ANGLE_INTO_A_LINE=60
• DBSCAN_DECREMENT_STEP=2 #by default
• DBSCAN_STRAIT_BASELINES= #-d to activate
• DBSCAN_FINAL_POINTS_PER_LINE=10 # final baseline points

• BaseLinPages is written by Moisés Pastor
• GT_Tool_PAGE is written by Jorge Martínez Vargas

• cd SRC
• make