{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# image processing\n",
    "### Average Hash\n",
    "- 이미지를 비교 가능한 해시 값으로 나타낸 것\n",
    "- 해시 함수 MD5, SHA256 등을 이용해 데이터 값을 간단한 해시 값으로 변환할 수 있음\n",
    "- 이미지가 비슷한지 등을 검출할 때는 해시함수를 사용하면 안됨. 해상도 크기 조정, 색조 보정, 압축 형식 변경 등으로 해시값이 달라짐"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2017-10-29T05:51:58.038255Z",
     "start_time": "2017-10-29T05:51:57.944823Z"
    },
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2017-10-29T05:53:17.470410Z",
     "start_time": "2017-10-29T05:53:17.460198Z"
    },
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def average_hash(fname, size = 16):\n",
    "    img = Image.open(fname)\n",
    "    img = img.convert('L') # 1을 지정하면 이진화, RGB, RGBA, CMYK 등의 모드도 지원\n",
    "    img = img.resize((size, size), Image.ANTIALIAS)\n",
    "    pixel_data = img.getdata()\n",
    "    pixels = np.array(pixel_data)\n",
    "    pixels = pixels.reshape((size, size))\n",
    "    avg = pixels.mean()\n",
    "    diff = 1 * (pixels > avg)\n",
    "    return diff"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2017-10-29T05:57:21.471308Z",
     "start_time": "2017-10-29T05:57:21.464335Z"
    },
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def np2hash(ahash):\n",
    "    bhash = []\n",
    "    for nl in ahash.tolist():\n",
    "        s1 = [str(i)for i in nl]\n",
    "        s2 = ''.join(s1)\n",
    "        i = int(s2, 2)\n",
    "        bhash.append('%04x' % i)\n",
    "    return ''.join(bhash)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2017-10-29T05:57:21.919154Z",
     "start_time": "2017-10-29T05:57:21.912923Z"
    }
   },
   "outputs": [],
   "source": [
    "ahash = average_hash('eiffel_tower.jpeg')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2017-10-29T05:57:22.130937Z",
     "start_time": "2017-10-29T05:57:22.126482Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 1]\n",
      " [0 0 1 1 1 1 1 1 0 0 0 0 0 0 1 1]\n",
      " [0 0 1 1 1 1 1 1 0 0 0 0 1 1 1 1]\n",
      " [0 1 1 1 1 1 1 1 0 0 0 0 1 1 1 1]\n",
      " [0 1 1 1 1 1 1 1 0 0 0 1 1 1 1 1]\n",
      " [1 1 1 1 1 1 1 0 0 1 1 1 1 1 1 1]\n",
      " [1 1 1 1 1 1 1 0 0 1 1 1 1 1 1 1]\n",
      " [1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1]\n",
      " [1 1 1 1 1 1 1 0 0 1 1 1 1 1 1 1]\n",
      " [1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1]\n",
      " [0 1 1 1 1 1 1 0 0 0 1 1 1 1 1 1]\n",
      " [0 0 1 1 1 1 1 0 0 0 1 1 1 1 0 0]\n",
      " [0 0 0 0 0 1 0 0 0 0 1 1 1 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]\n",
      " [0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0]]\n"
     ]
    }
   ],
   "source": [
    "print(ahash)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2017-10-29T05:57:22.682635Z",
     "start_time": "2017-10-29T05:57:22.679094Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7f813f033f0f7f0f7f1ffe7ffe7fff7ffe7fff7f7e3f3e3c0438008000000200\n"
     ]
    }
   ],
   "source": [
    "print(np2hash(ahash))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Caltech 101 데이터\n",
    "- 해밍 거리 : 같은 문자수를 가진 2개의 문자열에서 대응하는 위치에 있는 문자 중 다른 것의 개수\n",
    "    - 256글자의 해시값 중 얼마나 다른지 찾고 이 기반으로 이미지 차이를 구분"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2017-10-29T06:05:46.326719Z",
     "start_time": "2017-10-29T06:05:46.324248Z"
    },
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import os, re"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2017-10-29T06:20:49.952048Z",
     "start_time": "2017-10-29T06:20:08.396686Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.0 > ./image/101_ObjectCategories/chair/image_0016.jpg\n",
      "0.22265625 > ./image/101_ObjectCategories/stop_sign/image_0019.jpg\n",
      "0.2265625 > ./image/101_ObjectCategories/chair/image_0031.jpg\n",
      "0.23046875 > ./image/101_ObjectCategories/airplanes/image_0129.jpg\n",
      "0.234375 > ./image/101_ObjectCategories/umbrella/image_0009.jpg\n",
      "0.23828125 > ./image/101_ObjectCategories/airplanes/image_0124.jpg\n",
      "0.24609375 > ./image/101_ObjectCategories/chair/image_0001.jpg\n",
      "0.24609375 > ./image/101_ObjectCategories/chair/image_0002.jpg\n",
      "0.24609375 > ./image/101_ObjectCategories/dragonfly/image_0001.jpg\n",
      "ok\n"
     ]
    }
   ],
   "source": [
    "search_dir = \"./image/101_ObjectCategories/\"\n",
    "cache_dir = \"./image/cache_avhash\"\n",
    "\n",
    "if not os.path.exists(cache_dir):\n",
    "    os.mkdir(cache_dir)\n",
    "    \n",
    "def average_hash(fname, size = 16):\n",
    "    fname2 = fname[len(search_dir):]\n",
    "    # image cache\n",
    "    \n",
    "    cache_file = cache_dir + \"/\" + fname2.replace('/','_') + '.csv'\n",
    "    if not os.path.exists(cache_file):\n",
    "        img = Image.open(fname)\n",
    "        img = img.convert('L').resize((size, size), Image.ANTIALIAS)\n",
    "        pixels = np.array(img.getdata()).reshape((size,size))\n",
    "        avg = pixels.mean()\n",
    "        px = 1 * (pixels > avg)\n",
    "        np.savetxt(cache_file, px, fmt=\"%.0f\", delimiter=\",\")\n",
    "    else:\n",
    "        px = np.loadtxt(cache_file, delimiter=\",\")\n",
    "    return px\n",
    "\n",
    "def hamming_dist(a, b):\n",
    "    aa = a.reshape(1, -1)\n",
    "    ab = b.reshape(1, -1)\n",
    "    dist = (aa != ab).sum()\n",
    "    return dist\n",
    "\n",
    "def enum_all_files(path):\n",
    "    for root, dirs, files in os.walk(path):\n",
    "        for f in files:\n",
    "            fname = os.path.join(root, f)\n",
    "            if re.search(r'\\.(jpg|jpeg|pnp)$', fname):\n",
    "                yield fname\n",
    "                                \n",
    "def find_image(fname, rate):\n",
    "    src = average_hash(fname)\n",
    "    for fname in enum_all_files(search_dir):\n",
    "        dst = average_hash(fname)\n",
    "        diff_r = hamming_dist(src, dst) / 256\n",
    "        if diff_r < rate:\n",
    "            yield (diff_r, fname)\n",
    "            \n",
    "srcfile = search_dir + \"/chair/image_0016.jpg\"\n",
    "html = \"\"\n",
    "sim = list(find_image(srcfile, 0.25))\n",
    "sim = sorted(sim, key = lambda x: x[0])\n",
    "for r, f in sim:\n",
    "    print(r, \">\", f)\n",
    "    s = '<div style=\"float:left;\"><h3>[ 차이 : ' + str(r) + '-' + os.path.basename(f) + ']</h3>' + \\\n",
    "        '<p><a herf=\"' + f + '\"><img src=\"' + f + '\" width=400>' + '</a></p></div>'\n",
    "    html += s\n",
    "\n",
    "html = \"\"\"<html><head><meta charset=\"utf8\">/head>\n",
    "       <body><h3> 원래 이미지 </h3><p>\n",
    "       <img src = '{0}' width=400></p>{1}</body></html>\"\"\".format(srcfile, html)\n",
    "with open(\"./avgash-search-output.html\", \"w\", encoding=\"utf-8\") as f:\n",
    "    f.write(html)\n",
    "print(\"ok\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CNN\n",
    "- 일정한 크기로 리사이즈한 후, 24비트 RGB 형식으로 변환 -> Numpy 배열로 저장"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "import os, glob\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "# 분류 대상 카테고리 선택\n",
    "caltech_dir"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.1"
  },
  "latex_envs": {
   "LaTeX_envs_menu_present": true,
   "autocomplete": true,
   "bibliofile": "biblio.bib",
   "cite_by": "apalike",
   "current_citInitial": 1,
   "eqLabelWithNumbers": true,
   "eqNumInitial": 1,
   "hotkeys": {
    "equation": "Ctrl-E",
    "itemize": "Ctrl-I"
   },
   "labels_anchors": false,
   "latex_user_defs": false,
   "report_style_numbering": false,
   "user_envs_cfg": false
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}