{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "c0",
   "metadata": {},
   "source": [
    "# Indra OCR quickstart\n",
    "\n",
    "Extract text from an image file, and from an image already in memory. The first run downloads the models from Hugging Face (about 73 MB)."
   ]
  },
  {
   "cell_type": "code",
   "id": "c1",
   "metadata": {},
   "source": [
    "%pip install -q indra-ocr"
   ],
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "code",
   "id": "c2",
   "metadata": {},
   "source": [
    "import indra_ocr\n",
    "\n",
    "conv = indra_ocr.load(threads=4)"
   ],
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "c3",
   "metadata": {},
   "source": [
    "## An image file\n",
    "\n",
    "`convert` reads images (PNG, JPEG, TIFF, BMP, WebP) and PDFs. It yields `(page index, Page)`; a multi-page TIFF gives one page per frame."
   ]
  },
  {
   "cell_type": "code",
   "id": "c4",
   "metadata": {},
   "source": [
    "for i, page in conv.convert(\"scan.png\"):\n",
    "    print(page.markdown)   # headings, lists, tables as HTML\n",
    "    print(page.text)       # plain text, one line per line"
   ],
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "c5",
   "metadata": {},
   "source": [
    "## Lines, boxes and layout\n",
    "\n",
    "Every line has a box in image pixels, its text and a confidence, in reading order. `page.blocks` holds the layout blocks (label, box, text)."
   ]
  },
  {
   "cell_type": "code",
   "id": "c6",
   "metadata": {},
   "source": [
    "for ln in page.lines[:10]:\n",
    "    print(ln.box, round(ln.conf, 2), ln.text)\n",
    "\n",
    "for b in page.blocks[:5]:\n",
    "    print(b[\"label\"], b[\"bbox\"])"
   ],
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "c7",
   "metadata": {},
   "source": [
    "## An image in memory\n",
    "\n",
    "Pass a grayscale `uint8` numpy array to `conv.ocr`, for example a camera frame, a crop or a decoded upload."
   ]
  },
  {
   "cell_type": "code",
   "id": "c8",
   "metadata": {},
   "source": [
    "import cv2  # installed with indra-ocr\n",
    "\n",
    "color = cv2.imread(\"scan.png\")                 # or any uint8 BGR / RGB array\n",
    "gray = cv2.cvtColor(color, cv2.COLOR_BGR2GRAY)\n",
    "page = conv.ocr(gray)\n",
    "print(page.text)\n",
    "print(\"turned\", page.rotation, \"degrees, straightened\", page.skew, \"degrees\")"
   ],
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "id": "c9",
   "metadata": {},
   "source": [
    "## Math\n",
    "\n",
    "With `math=True`, lines with math come out as LaTeX (slower; the math models download on first use)."
   ]
  },
  {
   "cell_type": "code",
   "id": "c10",
   "metadata": {},
   "source": [
    "conv_math = indra_ocr.load(threads=4, math=True)\n",
    "for i, page in conv_math.convert(\"equations.png\"):\n",
    "    print(page.markdown)"
   ],
   "execution_count": null,
   "outputs": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}