#!/bin/bash
# Determines the medium of *-un.pdf (unknown) books empirically:
# OCRs two mid-book pages with the English model and counts common English
# stopwords — an Urdu page OCR'd as English yields almost none.
# Renames the file to -en.pdf or -ur.pdf accordingly.
set -u
ROOT="$(cd "$(dirname "$0")/.." && pwd)"
export TESSDATA_PREFIX="$ROOT/scripts/tessdata"
find "$ROOT" -name "*-un.pdf" -not -path "*/extracted-text/*" | sort | while read -r f; do
  pages=$(pdfinfo "$f" 2>/dev/null | awk '/^Pages:/{print $2}')
  mid=$(( ${pages:-20} / 2 ))
  tmp=$(mktemp -d)
  pdftoppm -gray -r 150 -f "$mid" -l $((mid+1)) "$f" "$tmp/pg" 2>/dev/null
  hits=0
  for img in "$tmp"/pg-*; do
    t=$(tesseract "$img" - -l eng --psm 3 2>/dev/null)
    h=$(echo "$t" | tr '[:upper:]' '[:lower:]' | grep -owE "the|and|of|in|to|is|are|was|which" | wc -l)
    hits=$((hits + h))
  done
  rm -rf "$tmp"
  if [ "$hits" -ge 10 ]; then new="${f%-un.pdf}-en.pdf"; med=English; else new="${f%-un.pdf}-ur.pdf"; med=Urdu; fi
  mv "$f" "$new"
  echo "$(basename "$f") -> $med (stopword hits: $hits)"
done