画像の類似度を比較するプログラムとそのユーティリティ

point_multi.sh

#!/bin/bash

set -eu

if [ $# -ne 2 ]; then
  echo "invalid arg" >&2
  exit 1
fi

for file in $2/*
do
  x=`./similarity.sh $1 $file`
  echo $1 $file $x
done

round_robin.sh

#!/bin/bash

set -eu

if [ $# -ne 1 ]; then
  echo "invalid arg" >&2
  exit 1
fi

files=(${1}/*)

for (( i = 0; i < ${#files[@]}; ++i ))
do
  for (( j = 0; j < ${#files[@]}; ++j ))
  do
    buf=`./similarity.sh  ${files[$i]} ${files[$j]}`
    echo ${files[$i]} ${files[$j]} ${buf}
  done
done

similarity.sh

#!/bin/sh
set -eu

help() {
  echo "$0 returns similarity of two given images."
  echo "they must be same width and height."
  echo ""
  echo "Usage:"
  echo "$0 image_file0 image_file1"
  exit 1
}

if [ $# -ne 2 ]; then
  help
fi

# check images to have same width and hight.
w1=`identify -format "%w" $1`
h1=`identify -format "%h" $1`
w2=`identify -format "%w" $2`
h2=`identify -format "%h" $2`

if [ $w1 -ne $w2 -o $h1 -ne $h2 ]; then
  help
fi

# generate tmporary file.
tmp1=$(mktemp "/tmp/${1##*/}.XXXXXX")
tmp2=$(mktemp "/tmp/${2##*/}.XXXXXX")

# when process finished, rm tmp files.
rm_tmps() {
  [ -e ${tmp1} ] && rm ${tmp1}
  [ -e ${tmp2} ] && rm ${tmp2}
}
trap rm_tmps INT TERM EXIT

# convert images to color histograms.
convert $1 -format '%c' histogram:info:- | sed 's/:.*#/ /' | awk '{print $2, $1}' > ${tmp1}
convert $2 -format '%c' histogram:info:- | sed 's/:.*#/ /' | awk '{print $2, $1}' > ${tmp2}

# summate minimums
sumofmins=`join ${tmp1} ${tmp2} | awk 'BEGIN{sum=0} {if($2 < $3) sum=sum+$2; else sum=sum+$3} END {print sum}'`

# normalize sumofmins
echo "scale=3; ${sumofmins} / (${w1} * ${h1})" | bc

take_pics.sh

#!/bin/sh

set -eu

if [ $# -ne 2 ]; then
  echo "Usage:"
  echo "${0} num_of_pics dest_dir"
  exit 1
fi

[ ! -d $2 ] && mkdir ${2}

for i in `seq 0 ${1}`; do
  fswebcam -r 640x480 --jpeg 85 -D 1 -S 35 --no-banner ${2}/`date +%Y-%m-%d-%H%M%S`.jpg
done

to_csv.py

# -*- conding:utf-8 -*-

import sys
import re

table = {}
column = []
row = []

sr = re.compile( ' +'  )
for line in open( sys.argv[1], 'r' ):
  l, r, v = sr.split( line.strip() )
  l = l.strip()
  r = r.strip()

  if not l in table:
    table[ l ] = {}

  table[ l ][ r ] = float( v )
  column.append( l )
  row.append( r )

column = sorted( list( set( column ) ) )
row    = sorted( list( set( row ) ) )

for r in row:
  print( ", ", r, end="")
print()
  
for c in column:
  print( c, end="" )
  for r in row:
    print( ", ",  table[ c ][ r ], end = "" )
  print()

similarity.sh img1 img2 : img1とimg2の類似度を計算する。
take_pics.sh num destdir : webカメラでnum枚の画像を取り、destdirに保存する。
round_robin.sh dir: dirに保存してある全ての画像に対して総当りで類似度を計算する。
to_csv.py txt: round_robin.shの出力結果を保存したファイルtxtをcsvに変換する。
point_multi.sh img dir : imgとdirにある全てのファイルの類似度を比較する。