# Last edited on 2010-08-09 17:10:21 by stolfilocal

INTRO

  This notebook documents the creation and processing of
  a test dataset for the "text-tracking" project,
  which includes the indetification, segmentation,
  tracking, and interpretation of text in videos.
  
  See 00-README for the structure and organization
  of the dataset.
  
CREATING THE DATA REPOSITORY
  
    forig=data/full/orig
    ftrue=data/full/true
    fmask=data/full/mask
    mkdir -p {${forig},${ftrue},${fmask}}
    
    See the "FETCH" subdirectory for documentation on the source
    and pre-processing of each image dataset.
  
>>> TO DO/REDO: >>>


CREATING THE TEXT+SURROUND MASKS

  For each image in I created a mask "full/mask/${NNN}.png" containing
  the text and some surround, by dilating the true mask.  The amount of dilation
  varied case-by-case:

    make-text-masks.sh

CREATING "PROPER SCALE" IMAGES

  Creating "natural scale" reduced versions of all images (where the
  strokes and interstroke gaps are about 1.42 pixels wide):

    ( cd data && split-text-images.sh )

    display data/crop/orig/*/*.png &
    display -filter box -resize '400%' data/cnat/orig/*/*.png &

    display data/crop/true/*/*.png &
    display -filter box -resize '400%' data/cnat/true/*/*.png &

    display data/crop/mask/*/*.png &
    display -filter box -resize '400%' data/cnat/mask/*/*.png &