Last edited on 2009-01-30 03:07:28 by stolfi

Processing the IAB-2002 fragment pictures
04 - Segmenting the images

======================================================================
% TO DO LATER
% 
% Before doing this step, check the segmentation images at
% UFF, and create touched-up versions of the -s.jpg images (-t.jpg) to
% fix bad segmentations. Also add extra seed points on the stage paper,
% on the mirror, and on the green monoliths.
  
UPDATING THE FRAGMENT NUMBERS AND POSITIONS

  Futher numbers were collected by HCGL's students at UFF
  in a different format.  Here is the conversion from UFF to Unicamp
  format: 

    pushd coords 
    cat fragcenters-uff-old.txt \
      | ../tools/convert-fragcenters-uff-unicamp \
        -f ../tools/read-table.gawk \
        -v table=../raw-to-sorted.tbl \
      > fragcenters-uff.txt
    popd

  This data should eventually be merged into fragcenters.txt,
  replacing the old invented fragment numbers.
  Here is the first try of that:
  
    pushd coords 
    cat fragcenters.txt \
      | ../tools/replace-fragment-numbers \
      > fragcenters-renumbered.txt
    popd
  
  The old "{F}.fctrs" and "{F}-f.jpg" files were renamed
  temporarily "{F}-o.fctrs" and "{F}-o-f.jpg". Then we 
  executed
  
    tools/write-fragcenters fragcenters-renumbered.txt
    
  The script creates files "{F}.fctrs" for each image "{F}.jpg" listed
  in "fragcenters.txt", and also a browse-size image "{F}-f.jpg" 
  showing the fragment centers and numbers.
  
  Comparing the two:
  
    find ./ -name '*-n.jpg' -print | grep stds | sort > .tmp
    display -title '%d/%f' `cat .tmp`
    
  Unfortunately the UFF data is too buggy and parctically useless.
  It did reveal some errors in the original data though.
  Undid the renaming.
  
  Checking whether there was any "{F}.fctrs" not generated by 
  the script:
  
    find ./ -name '*-o.fctrs' -print | sort > .ooo
    find ./ -name '*.fctrs' -print | sort > .ttt
    bool 1-2 .ttt .ooo > .nnn
    cat .ooo | sed -e 's:-o.fctrs:.fctrs:g' | sort > .sss
    diff -Bb .sss .nnn


SEGMENTING THE IMAGES

  Preliminary segmentation of the images is done by the script "segment-image"
  in the "tools" directory.  

  Collecting the names of images that are worth segmenting:
    
     find TEC/ PUB/ PRE/ \
         -type d \( -name 'stds' -o -name 'mirr' \) \
         -print \
       > .dirs
     find `cat .dirs` \
         -type f -name '*-s.jpg' \
         -print \
       | sed -e 's/-s[.]jpg//' \
       > .images
    
  Segmenting them: 

     foreach img ( `cat .images | sed -e 's:[#].*$::g'` )
       echo "=== ${img} ===================================="
       nice tools/segment-image ${img}
     end

COLLECTING ALL IMAGES OF ALL FRAGMENTS

  The goal here is to bring together all images of the same fragment,
  and display them in a format that makes it easy to read off the 
  official fragment number.

  Create table listing all extracted fragment images
  and the corresponding fragment names:

    find {PRE,TEC,ETC,PUB}/ \
        -name '*.seeds' \
        -print \
      > .seedfiles

    /bin/rm -f fragimage-to-fragname.tbl
    touch fragimage-to-fragname.tbl
    foreach sf ( `cat .seedfiles` )
      set img = `echo ${sf:r} | sed -e 's:-s$::g'`
      echo $sf $img
      cat $sf \
        | gawk -v img=${img} \
            ' /[=]/{ \
                num=$6; nam=$8; \
                if ((num+0) > 1) \
                  { printf "%s-seg-%s.png %s\n", img, num, nam; } \
              } ' \
        >> fragimage-to-fragname.tbl
    end

  Create list of fragments:

    cat fragimage-to-fragname.tbl \
       | gawk '//{print $2;}' \
       | sort | uniq \
       > .fragments

  Create the per-fragment directories:
    
    mkdir FRG
    foreach frg ( `cat .fragments` )
      mkdir FRG/${frg}
    end

  Now make, for each fragment, a list of all segment images 
  of that fragment:

    /bin/rm -f FRG/*/seg-images.dir
    cat fragimage-to-fragname.tbl \
      | sort -b +1 -2 \
      | gawk \
         ' BEGIN { ofrg = ""; } \
           // { img=$1; frg=$2; \
             if (frg != ofrg) \
               { if (ofrg != "") { close(ofile); } \
                 ofrg = frg; ofile = ("FRG/" frg "/seg-images.dir"); \
               } \
             printf "%s\n", img >> ofile; \
             next; \
           } \
           END { if (ofrg != "") { close(ofile); } } \
         '

EXTRACTING IMAGES OF INDIVIDUAL FRAGMENTS

  Extracting individual fragments:

     foreach img ( `cat .images | sed -e 's:[#].*$::g'` )
       echo "=== ${img} ===================================="
       nice tools/extract-segments ${img}
     end

  (This step should be improved. The generation of the full-scale
  fragment masks from the IFT label map takes too long and repeats the
  same work many times. Consider some special-purpose program to
  generate those masks.)

GATHERING ALL IMAGES OF EACH FRAGMENT

   Check whether we missed something:

    find {PRE,TEC,ETC,PUB}/ \
        -name '*-seg-[0-9][0-9][0-9].png' \
        -print \
      | sort  \
      > .segimages

    cat fragimage-to-fragname.tbl \
      | gawk ' //{print $1;} ' \
      | sort \
      > .expected-segimages

    bool 1-2 .expected-segimages .segimages > .missing-images
    bool 1-2 .segimages .expected-segimages > .spurious-images
    echo ' '; wc -l .expected-segimages .segimages .missing-images .spurious-images
    echo "spurious segmented fragment images: "
    cat .spurious-images

 Now merge all images of each fragment into a single image:

    foreach frg ( `cat .fragments` )
      tools/gather-fragment-images -show ${frg}
    end