ENPDataset_preprocess.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\users\\jhelu\\miniconda3\\envs\\cs682\\lib\\site-packages\\numpy\\_distributor_init.py:32: UserWarning: loaded more than 1 DLL from .libs:\n",
      "c:\\users\\jhelu\\miniconda3\\envs\\cs682\\lib\\site-packages\\numpy\\.libs\\libopenblas.PYQHXLVVQ7VESDPUVUADXEVJOBGHJPAY.gfortran-win_amd64.dll\n",
      "c:\\users\\jhelu\\miniconda3\\envs\\cs682\\lib\\site-packages\\numpy\\.libs\\libopenblas.TXA6YQSD3GCQQC22GEQ54J2UDCXDXHWN.gfortran-win_amd64.dll\n",
      "  stacklevel=1)\n"
     ]
    }
   ],
   "source": [
    "import xml.dom.minidom\n",
    "import numpy as np\n",
    "import os\n",
    "import json\n",
    "# from PIL import Image\n",
    "from skimage import io, transform, util\n",
    "import matplotlib.pyplot as plt\n",
    "from scipy.signal import argrelextrema, argrelmin, argrelmax\n",
    "from scipy.signal import savgol_filter\n",
    "from skimage.draw import polygon\n",
    "from skimage.color import rgb2gray\n",
    "\n",
    "def consecutive(data, stepsize=1):\n",
    "    return np.split(data, np.where(np.diff(data) != stepsize)[0]+1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "file_list = os.listdir(\"datasets/prima\")\n",
    "ground_truth = {}\n",
    "for file_name in file_list:\n",
    "    if file_name[:3] == 'pc-' and file_name[-4:] == '.xml':\n",
    "#         print(file_name)\n",
    "        ground_truth[file_name] = {}\n",
    "#         file_name = \"datasets/prima/pc-00008061.xml\"\n",
    "        doc = xml.dom.minidom.parse(\"datasets/prima/\"+file_name)        \n",
    "        for txt_reg in doc.getElementsByTagName(\"TextRegion\"):\n",
    "            if txt_reg.getAttribute(\"type\") == 'paragraph':# and txt_reg.getAttribute(\"id\") in r[3:4]:# == 'r116': \n",
    "\n",
    "                for text in txt_reg.getElementsByTagName(\"TextEquiv\"):\n",
    "                    text_str = text.getElementsByTagName(\"Unicode\")[0].firstChild.data\n",
    "                    lines = text_str.splitlines()\n",
    "                    word_list = []\n",
    "                    for line in lines:\n",
    "                        word_list.append(line.split(' '))\n",
    "                    if len(word_list) > 2:\n",
    "#                         print(txt_reg.getAttribute(\"type\"), txt_reg.getAttribute(\"id\"))\n",
    "                        ground_truth[file_name][txt_reg.getAttribute(\"id\")] = word_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "count_lines =  sum(len(ground_truth[k1][k2]) for k1 in ground_truth for k2 in ground_truth[k1] )\n",
    "count_words = sum(len(word_list) for k1 in ground_truth for k2 in ground_truth[k1] for word_list in ground_truth[k1][k2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "def process_images(ground_truth, hor_f, hor_d, vert_f, vert_d, vert_c_p):\n",
    "    images = {}\n",
    "    incorrect_lines = {}\n",
    "    incorrect_words = {}\n",
    "    for xml_file in ground_truth:\n",
    "    #     if xml_file !=\"pc-00008061.xml\":\n",
    "    #         continue\n",
    "        #parse sml doc\n",
    "        doc = xml.dom.minidom.parse(\"datasets/prima/\"+xml_file)\n",
    "        images[xml_file] = {}\n",
    "        incorrect_lines[xml_file] = {}\n",
    "        incorrect_words[xml_file] = {}\n",
    "        #get image name from xml doc\n",
    "        for a in doc.getElementsByTagName(\"Page\"):\n",
    "            image_name = a.getAttribute('imageFilename')\n",
    "\n",
    "        im_full = rgb2gray(io.imread(\"datasets/prima/\"+image_name))\n",
    "#         print(\"image name \", image_name)\n",
    "        im_full = np.asarray(im_full)\n",
    "        #select regions from image with multiple lines (>2), already present in groundtruth dict\n",
    "        for txt_reg in doc.getElementsByTagName(\"TextRegion\"):\n",
    "            if txt_reg.getAttribute(\"type\") == 'paragraph' and txt_reg.getAttribute(\"id\") in ground_truth[xml_file].keys():# == 'r116': \n",
    "    #             if txt_reg.getAttribute(\"id\")!= 'r108':\n",
    "    #                 continue\n",
    "                # read bounding points of region\n",
    "                x_y = []\n",
    "                for coords in txt_reg.getElementsByTagName(\"Coords\"):\n",
    "                    for point in coords.getElementsByTagName(\"Point\"):\n",
    "                        x_y.append([point.getAttribute(\"x\"), point.getAttribute(\"y\")])\n",
    "                x_y = np.asarray(x_y).astype(int)\n",
    "                #create mask to crop out layer\n",
    "                mask = np.zeros((max(x_y[:,1])-min(x_y[:,1]), max(x_y[:,0])-min(x_y[:,0])), dtype=np.uint8)\n",
    "                rr, cc = polygon(x_y[:,1] - min(x_y[:,1]), x_y[:,0] - min(x_y[:,0]))\n",
    "                mask[rr, cc] = 1\n",
    "    #             plt.figure()\n",
    "    #             plt.imshow(mask, cmap='gray')\n",
    "                #crop region based on min-max of bounding points\n",
    "                im = im_full[ min(x_y[:,1]): max(x_y[:,1]), min(x_y[:,0]):max(x_y[:,0])] \n",
    "                im = util.invert(im) * mask\n",
    "                im = util.invert(im)\n",
    "                #get horizontal proj. to segment line\n",
    "                hor_proj = np.sum(util.invert(im), axis=1)\n",
    "                hor_proj_smooth = savgol_filter(hor_proj, hor_f, hor_d)\n",
    "    #             print(min(hor_proj), max(hor_proj))\n",
    "    #             plt.figure(figsize=(30,5))\n",
    "    #             plt.plot(np.arange(max(x_y[:,1])-min(x_y[:,1])), hor_proj)\n",
    "    #             plt.plot(np.arange(max(x_y[:,1])-min(x_y[:,1])), hor_proj_smooth, c='r')\n",
    "    #             plt.show()\n",
    "                #get local minima from the smoothened hor. proj. \n",
    "                index_maxima = argrelextrema(hor_proj_smooth, np.greater)[0]\n",
    "                index_minima = argrelextrema(hor_proj_smooth, np.less)[0]\n",
    "\n",
    "#                 plt.figure()\n",
    "#                 plt.imshow(im, cmap='gray')\n",
    "#                 plt.scatter(x = x_y[:,0]-min(x_y[:,0]), y = x_y[:,1] - min(x_y[:,1]))\n",
    "#                 for ind in index_minima:\n",
    "#                     plt.scatter(x = 0, y=ind, c='r')\n",
    "#                 plt.show()\n",
    "                image_list = []\n",
    "                line_number = 0\n",
    "                for i in range(1, len(index_minima)):\n",
    "                    image_line = []\n",
    "                    #crop out each line\n",
    "                    img_crop = im[index_minima[i-1]:index_minima[i], : ]\n",
    "                    if np.sum(img_crop) <= 10 or np.sum(util.invert(img_crop)) <= 10:\n",
    "                        continue\n",
    "#                     plt.figure()\n",
    "#                     plt.imshow(img_crop, cmap='gray')\n",
    "#                     plt.show()\n",
    "                    #get vert. proj. to segment words\n",
    "                    vert_proj = np.sum(util.invert(img_crop), axis=0)\n",
    "                    vert_proj_smooth = savgol_filter(vert_proj, vert_f, vert_d)\n",
    "    #                 plt.figure(figsize=(30,5))\n",
    "    #                 plt.plot(np.arange(vert_proj.shape[0]), vert_proj)\n",
    "    #                 plt.plot(np.arange(vert_proj.shape[0]), vert_proj_smooth, c='r')\n",
    "    #                 plt.show()\n",
    "                    index_min = argrelextrema(vert_proj_smooth, np.less)[0]\n",
    "        #             print(len(np.where(vert_proj_smooth<=0)[0]), len(index_min))\n",
    "                    #get sets of consecutive inidces where vert_proj[index] == 0, represents spaces\n",
    "                    spaces = consecutive(np.where(vert_proj_smooth<=0)[0], vert_c_p)\n",
    "    #                 fig, axes = plt.subplots(1, len(spaces)-1, figsize=(30, 2))\n",
    "    #                 for _i, ax in enumerate(axes.flat): # ax in range(1,len(spaces)):\n",
    "                    for _i in range(len(spaces)-1):\n",
    "                        #crop out a word\n",
    "                        word = img_crop[:, spaces[_i][int(len(spaces[_i])/2)]: spaces[_i+1][int(len(spaces[_i+1])/2)]]\n",
    "                        if np.sum(util.invert(word)) <=5 or np.sum(word) <= 5:\n",
    "                            continue\n",
    "    #                     ax.imshow(word, cmap='gray')\n",
    "                        image_line.append(word)\n",
    "                    image_list.append(image_line)\n",
    "                    try:\n",
    "                        if len(ground_truth[xml_file][txt_reg.getAttribute(\"id\")][line_number]) != len(image_line):\n",
    "#                             print(\"incorrect no of words \", xml_file, txt_reg.getAttribute(\"id\"))\n",
    "                            if txt_reg.getAttribute(\"id\") not in incorrect_words[xml_file]:\n",
    "                                incorrect_words[xml_file][txt_reg.getAttribute(\"id\")] = 1\n",
    "                            else:\n",
    "                                incorrect_words[xml_file][txt_reg.getAttribute(\"id\")] += 1\n",
    "                        line_number += 1\n",
    "                    except IndexError:\n",
    "                        pass\n",
    "                        #print(\"incorrect number of lines\", xml_file, txt_reg.getAttribute(\"id\"), len(ground_truth[xml_file][txt_reg.getAttribute(\"id\")]))\n",
    "                images[xml_file][txt_reg.getAttribute(\"id\")] = image_list\n",
    "                if len(ground_truth[xml_file][txt_reg.getAttribute(\"id\")]) != len(image_list):\n",
    "                    if txt_reg.getAttribute(\"id\") not in incorrect_lines[xml_file]:\n",
    "                        incorrect_lines[xml_file][txt_reg.getAttribute(\"id\")] = 1\n",
    "                    else:\n",
    "                        incorrect_lines[xml_file][txt_reg.getAttribute(\"id\")] += 1\n",
    "                        \n",
    "    return images, incorrect_lines, incorrect_words"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 0 0 0 0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\users\\jhelu\\miniconda3\\envs\\cs682\\lib\\site-packages\\ipykernel_launcher.py:17: FutureWarning: The behavior of rgb2gray will change in scikit-image 0.19. Currently, rgb2gray allows 2D grayscale image to be passed as inputs and leaves them unmodified as outputs. Starting from version 0.19, 2D arrays will be treated as 1D images with 3 channels.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10.489013257488876\n",
      "10.537328663070335\n",
      "0 0 0 0 1\n",
      "10.489013257488876\n",
      "10.537042026634257\n",
      "0 0 0 0 2\n",
      "10.489013257488876\n",
      "10.537328663070335\n",
      "0 0 0 0 3\n",
      "10.489013257488876\n",
      "10.536755390198179\n",
      "0 0 0 0 4\n",
      "10.489013257488876\n",
      "10.535035571581718\n",
      "0 0 0 1 0\n",
      "10.489013257488876\n",
      "10.524716659882937\n",
      "0 0 0 1 1\n",
      "10.489013257488876\n",
      "10.52557656919117\n",
      "0 0 0 1 2\n",
      "10.489013257488876\n",
      "10.524143387010783\n",
      "0 0 0 1 3\n",
      "10.489013257488876\n",
      "10.524716659882937\n",
      "0 0 0 1 4\n",
      "10.489013257488876\n",
      "10.52328347770255\n",
      "0 0 0 2 0\n",
      "10.489013257488876\n",
      "10.582330583534457\n",
      "0 0 0 2 1\n",
      "10.489013257488876\n",
      "10.582330583534457\n",
      "0 0 0 2 2\n",
      "10.489013257488876\n",
      "10.581757310662303\n",
      "0 0 0 2 3\n",
      "10.489013257488876\n",
      "10.580610764917992\n",
      "0 0 0 2 4\n",
      "10.489013257488876\n",
      "10.579177582737607\n",
      "0 0 1 0 0\n",
      "10.489013257488876\n",
      "10.428406817360996\n",
      "0 0 1 0 1\n",
      "10.489013257488876\n",
      "10.428406817360996\n",
      "0 0 1 0 2\n",
      "10.489013257488876\n",
      "10.42812018092492\n",
      "0 0 1 0 3\n",
      "10.489013257488876\n",
      "10.42812018092492\n",
      "0 0 1 0 4\n",
      "10.489013257488876\n",
      "10.428693453797074\n",
      "0 0 1 1 0\n",
      "10.489013257488876\n",
      "10.408915539707746\n",
      "0 0 1 1 1\n",
      "10.489013257488876\n",
      "10.409202176143822\n",
      "0 0 1 1 2\n",
      "10.489013257488876\n",
      "10.40862890327167\n",
      "0 0 1 1 3\n",
      "10.489013257488876\n",
      "10.408342266835591\n",
      "0 0 1 1 4\n",
      "10.489013257488876\n",
      "10.408915539707746\n",
      "0 0 1 2 0\n",
      "10.489013257488876\n",
      "10.463376462562415\n",
      "0 0 1 2 1\n",
      "10.489013257488876\n",
      "10.462803189690261\n",
      "0 0 1 2 2\n",
      "10.489013257488876\n",
      "10.462516553254183\n",
      "0 0 1 2 3\n",
      "10.489013257488876\n",
      "10.463089826126337\n",
      "0 0 1 2 4\n",
      "10.489013257488876\n",
      "10.463376462562415\n",
      "0 0 2 0 0\n",
      "10.489013257488876\n",
      "10.462803189690261\n",
      "0 0 2 0 1\n",
      "10.489013257488876\n",
      "10.462516553254183\n",
      "0 0 2 0 2\n",
      "10.489013257488876\n",
      "10.462803189690261\n",
      "0 0 2 0 3\n",
      "10.489013257488876\n",
      "10.462516553254183\n",
      "0 0 2 0 4\n",
      "10.489013257488876\n",
      "10.463089826126337\n",
      "0 0 2 1 0\n",
      "10.489013257488876\n",
      "10.431846454593922\n",
      "0 0 2 1 1\n",
      "10.489013257488876\n",
      "10.432419727466076\n",
      "0 0 2 1 2\n",
      "10.489013257488876\n",
      "10.431846454593922\n",
      "0 0 2 1 3\n",
      "10.489013257488876\n",
      "10.431273181721767\n",
      "0 0 2 1 4\n",
      "10.489013257488876\n",
      "10.43213309103\n",
      "0 0 2 2 0\n",
      "10.489013257488876\n",
      "10.383404896896874\n",
      "0 0 2 2 1\n",
      "10.489013257488876\n",
      "10.383118260460797\n",
      "0 0 2 2 2\n",
      "10.489013257488876\n",
      "10.383978169769028\n",
      "0 0 2 2 3\n",
      "10.489013257488876\n",
      "10.38283162402472\n",
      "0 0 2 2 4\n",
      "10.489013257488876\n",
      "10.38369153333295\n",
      "0 0 3 0 0\n",
      "10.489013257488876\n",
      "10.50694520084615\n",
      "0 0 3 0 1\n",
      "10.489013257488876\n",
      "10.507231837282228\n",
      "0 0 3 0 2\n",
      "10.489013257488876\n",
      "10.507231837282228\n",
      "0 0 3 0 3\n",
      "10.489013257488876\n",
      "10.507231837282228\n",
      "0 0 3 0 4\n",
      "10.489013257488876\n",
      "10.507518473718305\n",
      "0 0 3 1 0\n",
      "10.489013257488876\n",
      "10.486880650320746\n",
      "0 0 3 1 1\n",
      "10.489013257488876\n",
      "10.486020741012513\n",
      "0 0 3 1 2\n",
      "10.489013257488876\n",
      "10.486020741012513\n",
      "0 0 3 1 3\n",
      "10.489013257488876\n",
      "10.486020741012513\n",
      "0 0 3 1 4\n",
      "10.489013257488876\n",
      "10.486307377448592\n",
      "0 0 3 2 0\n",
      "10.489013257488876\n",
      "10.362480437063237\n",
      "0 0 3 2 1\n",
      "10.489013257488876\n",
      "10.362767073499315\n",
      "0 0 3 2 2\n",
      "10.489013257488876\n",
      "10.362480437063237\n",
      "0 0 3 2 3\n",
      "10.489013257488876\n",
      "10.36219380062716\n",
      "0 0 3 2 4\n",
      "10.489013257488876\n",
      "10.362480437063237\n",
      "0 0 4 0 0\n",
      "10.489013257488876\n",
      "10.56341257875336\n",
      "0 0 4 0 1\n",
      "10.489013257488876\n",
      "10.56341257875336\n",
      "0 0 4 0 2\n",
      "10.489013257488876\n",
      "10.56341257875336\n",
      "0 0 4 0 3\n",
      "10.489013257488876\n",
      "10.56341257875336\n",
      "0 0 4 0 4\n",
      "10.489013257488876\n",
      "10.563125942317283\n",
      "0 0 4 1 0\n",
      "10.489013257488876\n",
      "10.5490807569495\n",
      "0 0 4 1 1\n",
      "10.489013257488876\n",
      "10.5490807569495\n",
      "0 0 4 1 2\n",
      "10.489013257488876\n",
      "10.5490807569495\n",
      "0 0 4 1 3\n",
      "10.489013257488876\n",
      "10.5490807569495\n",
      "0 0 4 1 4\n",
      "10.489013257488876\n",
      "10.548794120513422\n",
      "0 0 4 2 0\n",
      "10.489013257488876\n",
      "10.398883264445043\n",
      "0 0 4 2 1\n",
      "10.489013257488876\n",
      "10.398596628008965\n",
      "0 0 4 2 2\n",
      "10.489013257488876\n",
      "10.39802335513681\n",
      "0 0 4 2 3\n",
      "10.489013257488876\n",
      "10.398596628008965\n",
      "0 0 4 2 4\n",
      "10.489013257488876\n",
      "10.398309991572889\n",
      "0 1 0 0 0\n",
      "10.266525987430617\n",
      "11.983696119515928\n",
      "0 1 0 0 1\n",
      "10.266525987430617\n",
      "11.983696119515928\n",
      "0 1 0 0 2\n",
      "10.266525987430617\n",
      "11.983982755952006\n",
      "0 1 0 0 3\n",
      "10.266525987430617\n",
      "11.98254957377162\n",
      "0 1 0 0 4\n",
      "10.266525987430617\n",
      "11.983409483079852\n",
      "0 1 0 1 0\n",
      "10.266525987430617\n",
      "11.97739011792223\n",
      "0 1 0 1 1\n",
      "10.266525987430617\n",
      "11.977676754358308\n",
      "0 1 0 1 2\n",
      "10.266525987430617\n",
      "11.977676754358308\n",
      "0 1 0 1 3\n",
      "10.266525987430617\n",
      "11.976243572177921\n",
      "0 1 0 1 4\n",
      "10.266525987430617\n",
      "11.97739011792223\n",
      "0 1 0 2 0\n",
      "10.266525987430617\n",
      "12.011499853815417\n",
      "0 1 0 2 1\n",
      "10.266525987430617\n",
      "12.013506308867958\n",
      "0 1 0 2 2\n",
      "10.266525987430617\n",
      "12.012933035995804\n",
      "0 1 0 2 3\n",
      "10.266525987430617\n",
      "12.01321967243188\n",
      "0 1 0 2 4\n",
      "10.266525987430617\n",
      "12.011786490251493\n",
      "0 1 1 0 0\n",
      "10.266525987430617\n",
      "11.850123540303949\n",
      "0 1 1 0 1\n",
      "10.266525987430617\n",
      "11.849263630995718\n",
      "0 1 1 0 2\n",
      "10.266525987430617\n",
      "11.849263630995718\n",
      "0 1 1 0 3\n",
      "10.266525987430617\n",
      "11.849550267431795\n",
      "0 1 1 0 4\n",
      "10.266525987430617\n",
      "11.849550267431795\n",
      "0 1 1 1 0\n",
      "10.266525987430617\n",
      "11.831205535522853\n",
      "0 1 1 1 1\n",
      "10.266525987430617\n",
      "11.830058989778545\n",
      "0 1 1 1 2\n",
      "10.266525987430617\n",
      "11.830058989778545\n",
      "0 1 1 1 3\n",
      "10.266525987430617\n",
      "11.829772353342467\n",
      "0 1 1 1 4\n",
      "10.266525987430617\n",
      "11.831492171958931\n",
      "0 1 1 2 0\n",
      "10.266525987430617\n",
      "11.875347546678745\n",
      "0 1 1 2 1\n",
      "10.266525987430617\n",
      "11.87563418311482\n",
      "0 1 1 2 2\n",
      "10.266525987430617\n",
      "11.874201000934436\n",
      "0 1 1 2 3\n",
      "10.266525987430617\n",
      "11.875347546678745\n",
      "0 1 1 2 4\n",
      "10.266525987430617\n",
      "11.87563418311482\n",
      "0 1 2 0 0\n",
      "10.266525987430617\n",
      "11.8696148179572\n",
      "0 1 2 0 1\n",
      "10.266525987430617\n",
      "11.8696148179572\n",
      "0 1 2 0 2\n",
      "10.266525987430617\n",
      "11.869901454393277\n",
      "0 1 2 0 3\n",
      "10.266525987430617\n",
      "11.869328181521121\n",
      "0 1 2 0 4\n",
      "10.266525987430617\n",
      "11.870188090829354\n",
      "0 1 2 1 0\n",
      "10.266525987430617\n",
      "11.835505082064012\n",
      "0 1 2 1 1\n",
      "10.266525987430617\n",
      "11.834931809191858\n",
      "0 1 2 1 2\n",
      "10.266525987430617\n",
      "11.834931809191858\n",
      "0 1 2 1 3\n",
      "10.266525987430617\n",
      "11.834358536319703\n",
      "0 1 2 1 4\n",
      "10.266525987430617\n",
      "11.836078354936166\n",
      "0 1 2 2 0\n",
      "10.266525987430617\n",
      "11.778464431284648\n",
      "0 1 2 2 1\n",
      "10.266525987430617\n",
      "11.778464431284648\n",
      "0 1 2 2 2\n",
      "10.266525987430617\n",
      "11.778751067720725\n",
      "0 1 2 2 3\n",
      "10.266525987430617\n",
      "11.777317885540338\n",
      "0 1 2 2 4\n",
      "10.266525987430617\n",
      "11.779610977028957\n",
      "0 1 3 0 0\n",
      "10.266525987430617\n",
      "11.893405642151608\n",
      "0 1 3 0 1\n",
      "10.266525987430617\n",
      "11.89311900571553\n",
      "0 1 3 0 2\n",
      "10.266525987430617\n",
      "11.893405642151608\n",
      "0 1 3 0 3\n",
      "10.266525987430617\n",
      "11.89311900571553\n",
      "0 1 3 0 4\n",
      "10.266525987430617\n",
      "11.893405642151608\n",
      "0 1 3 1 0\n",
      "10.266525987430617\n",
      "11.869328181521121\n",
      "0 1 3 1 1\n",
      "10.266525987430617\n",
      "11.869328181521121\n",
      "0 1 3 1 2\n",
      "10.266525987430617\n",
      "11.869901454393277\n",
      "0 1 3 1 3\n",
      "10.266525987430617\n",
      "11.8696148179572\n",
      "0 1 3 1 4\n",
      "10.266525987430617\n",
      "11.869328181521121\n",
      "0 1 3 2 0\n",
      "10.266525987430617\n",
      "11.731456055767984\n",
      "0 1 3 2 1\n"
     ]
    }
   ],
   "source": [
    "#incorrect words and lines\n",
    "hor_filter_size = [11, 21, 31]\n",
    "hor_function_degree = [1, 2]\n",
    "vert_filter_size = [11, 21, 25, 31, 41]\n",
    "vert_function_degree = [2, 3, 4]\n",
    "vert_consecutive_pixels = [3, 5, 7, 9, 10]\n",
    "\n",
    "incorrect_lines = np.zeros((4,2,7,3,5))\n",
    "incorrect_words = np.zeros((4,2,7,3,5))\n",
    "for h1 in range(len(hor_filter_size)):\n",
    "    for h2 in range(len(hor_function_degree)):\n",
    "        for v1 in range(len(vert_filter_size)):\n",
    "            for v2 in range(len(vert_function_degree)):\n",
    "                for v3 in range(len(vert_consecutive_pixels)):\n",
    "                    print(h1, h2, v1, v2, v3)\n",
    "                    _, e1, e2  = process_images(ground_truth, hor_filter_size[h1], hor_function_degree[h2],\n",
    "                                                vert_filter_size[v1], vert_function_degree[v2], vert_consecutive_pixels[v3])\n",
    "                    incorrect_lines[h1, h2, v1, v2, v3] = sum(e1[k1][k2] for k1 in e1 for k2 in e1[k1] )\n",
    "                    incorrect_words[h1, h2, v1, v2, v3] = sum(e2[k1][k2] for k1 in e2 for k2 in e2[k1] )\n",
    "                    print(sum(e1[k1][k2] for k1 in e1 for k2 in e1[k1] )/ count_lines *100)\n",
    "                    print(sum(e2[k1][k2] for k1 in e2 for k2 in e2[k1] )/ count_words *100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'hor_proj_smooth' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[1;32m<ipython-input-9-fc0def8edfda>\u001b[0m in \u001b[0;36m<module>\u001b[1;34m\u001b[0m\n\u001b[1;32m----> 1\u001b[1;33m \u001b[0mgrad\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mgradient\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mhor_proj_smooth\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m      2\u001b[0m \u001b[0mhor_proj_sort\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0msorted\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mhor_proj\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      3\u001b[0m \u001b[1;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[1;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;36m1\u001b[0m\u001b[1;33m,\u001b[0m\u001b[0mlen\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mgrad\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m-\u001b[0m\u001b[1;36m1\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      4\u001b[0m     \u001b[1;32mif\u001b[0m \u001b[0mgrad\u001b[0m\u001b[1;33m[\u001b[0m\u001b[0mi\u001b[0m\u001b[1;33m-\u001b[0m\u001b[1;36m1\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m<=\u001b[0m\u001b[1;36m0\u001b[0m \u001b[1;32mand\u001b[0m \u001b[0mgrad\u001b[0m\u001b[1;33m[\u001b[0m\u001b[0mi\u001b[0m\u001b[1;33m]\u001b[0m \u001b[1;33m==\u001b[0m\u001b[1;36m0\u001b[0m \u001b[1;32mand\u001b[0m \u001b[0mgrad\u001b[0m\u001b[1;33m[\u001b[0m\u001b[0mi\u001b[0m\u001b[1;33m+\u001b[0m\u001b[1;36m1\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m>\u001b[0m\u001b[1;36m0\u001b[0m \u001b[1;32mand\u001b[0m \u001b[0mhor_proj\u001b[0m\u001b[1;33m[\u001b[0m\u001b[0mi\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m<\u001b[0m \u001b[0mhor_proj_sort\u001b[0m\u001b[1;33m[\u001b[0m\u001b[0mint\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;36m0.2\u001b[0m\u001b[1;33m*\u001b[0m\u001b[0mlen\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mhor_proj\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      5\u001b[0m         \u001b[0mprint\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;34m\"minima \"\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mi\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mhor_proj\u001b[0m\u001b[1;33m[\u001b[0m\u001b[0mi\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mNameError\u001b[0m: name 'hor_proj_smooth' is not defined"
     ]
    }
   ],
   "source": [
    "grad = np.gradient(hor_proj_smooth)\n",
    "hor_proj_sort = sorted(hor_proj)\n",
    "for i in range(1,len(grad)-1):\n",
    "    if grad[i-1]<=0 and grad[i] ==0 and grad[i+1]>0 and hor_proj[i]< hor_proj_sort[int(0.2*len(hor_proj))]:\n",
    "        print(\"minima \", i, hor_proj[i])\n",
    "    elif grad[i] ==0 and grad[i+1]<0 and hor_proj[i]> hor_proj_sort[-int(0.2*len(hor_proj))]:\n",
    "        print(\"maxima\", i, hor_proj[i])\n",
    "grad_2 = np.gradient(grad)/max(x_y[:,1])\n",
    "#print(argrelextrema(grad_2, np.greater)[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7\n",
      "11\n",
      "11\n",
      "14\n",
      "11\n",
      "10\n",
      "10\n",
      "1\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x20a23bc66a0>"
      ]
     },
     "execution_count": 175,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(len(ground_truth['pc-00008061.xml']['r112']))\n",
    "for item in ground_truth['pc-00008061.xml']['r112']:\n",
    "    print(len(item))\n",
    "image = io.imread('datasets/prima/00008061.tif')\n",
    "plt.figure()\n",
    "plt.imshow(image, cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "images[]"
   ]
  }
 ],
 "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": "3.6.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}