pdf-to-hocr: fix ocr_line bug and add scaler

This commit fixes a bug where ocr_line elements would not have any
title for some PDFs (like ones created by OCRMyPDF).

This commit also adds PDF Metadata JSON as a requirement to make the
hOCR files, using the information contained within to estimate the DPI
and to scale the hOCR coordinates.

bin/pdf-to-hocr

@@ -2,6 +2,7 @@
 
 import sys
 import argparse
+import json
 
 from lxml import etree
 
@@ -121,7 +122,7 @@ def pdf_baseline_from_charboxes(charboxes):
     return float(m), int(c)
 
 
-def pdf_process_text_block(pageno, block, parent_block):
+def pdf_process_text_block(pageno, block, parent_block, scaler=lambda x: x):
     parelem = etree.Element('p', attrib={'class': 'ocr_par'})
 
     parelem.attrib['id'] = get_id(pageno, 'par')
@@ -137,11 +138,11 @@ def pdf_process_text_block(pageno, block, parent_block):
         lineelem = etree.Element('span', attrib={'class': 'ocr_line'})
 
         kv = {}
-        kv['bbox'] = [str(int(x)) for x in line['bbox']]
+        kv['bbox'] = get_bbox(line['bbox'], scaler)
 
-        last_wordelem, cboxes = pdf_process_characters(pageno, line, lineelem)
+        cboxes = pdf_process_characters(pageno, line, lineelem, scaler)
 
-        if last_wordelem is not None:
+        if cboxes:
             char_boxes += cboxes
 
             # TODO: Just use the dictionary properties?
@@ -151,17 +152,20 @@ def pdf_process_text_block(pageno, block, parent_block):
             lineelem.attrib['title'] = assemble_hocr_title_element(kv)
             lineelem.attrib['id'] = get_id(pageno, 'line')
 
-            if last_wordelem is not None:
-                lineelem.append(last_wordelem)
-
-        parelem.append(lineelem)
+            parelem.append(lineelem)
 
     kv = {}
-    kv = {'bbox': [str(int(x)) for x in block['bbox']]}
+    kv = {'bbox': get_bbox(block['bbox'], scaler)}
     parelem.attrib['title'] = assemble_hocr_title_element(kv)
     parent_block.append(parelem)
 
 
+def get_bbox(bbox_data, scaler):
+    d = [int(scaler(x)) for x in bbox_data]
+    d = [x if x > 0 else 0 for x in d]
+    d = [str(x) for x in d]
+    return d
+
 def _gather_word_data(pageno, wordelem, wordchar_bboxes, span):
     # Turn these calculations into a function
     word_bbox = [
@@ -179,8 +183,11 @@ def _gather_word_data(pageno, wordelem, wordchar_bboxes, span):
     wordelem.attrib['title'] = \
             assemble_hocr_title_element(word_data)
 
-
-def pdf_process_characters(pageno, line, lineelem):
+def pdf_process_characters(pageno, line, lineelem, scaler):
+    """
+    Process spans in a line, searching for spaces so that we can split the
+    characters into words.
+    """
     wordelem = etree.Element('span', attrib={'class': 'ocrx_word'})
     charelem = None
     word_start = True
@@ -221,8 +228,8 @@ def pdf_process_characters(pageno, line, lineelem):
 
             conf = float(100)
 
-            bbox = [int(x) for x in char['bbox']]
-            #bbox = [str(int(x)) for x in char['bbox']]
+            bbox = get_bbox(char['bbox'], scaler)
+            bbox = [int(x) for x in bbox] # temporary go back to int
             wordchar_bboxes.append(bbox)
             all_wordchar_bboxes.append(bbox)
 
@@ -240,21 +247,26 @@ def pdf_process_characters(pageno, line, lineelem):
         #    wordelem[-1].text += chr(0x200e)
 
     # Sometimes we find no legit chars in a word, in which case charelem is None
-    if found_any and len(wordchar_bboxes):
-        _gather_word_data(pageno, wordelem, wordchar_bboxes, span)
+    if found_any:
+        if not word_start:
+            lineelem.append(wordelem)
+            _gather_word_data(pageno, wordelem, wordchar_bboxes, span)
 
-        return wordelem, all_wordchar_bboxes
+        return all_wordchar_bboxes
 
-    return None, None
+    return None
 
 
-def pdf_page_to_hocr_page(page, pageno=None):
+def pdf_page_to_hocr_page(page, page_metadata, pageno=None):
     pagedata = page.get_text(option='rawdict')
     # TODO: left to right, right to left
 
+
+    width, height = page_metadata['estimated_default_render_res'][2:]
+
     kv = {}
-    w = str(int(pagedata['width']))
-    h = str(int(pagedata['height']))
+    w = str(int(width))
+    h = str(int(height))
     kv['bbox'] = ['0', '0', w, h]
 
     if pageno is not None:
@@ -262,10 +274,13 @@ def pdf_page_to_hocr_page(page, pageno=None):
     else:
         kv['ppageno'] = '0'
 
-    kv['image'] = 'https://archive.org/todo' # TODO: some image path?
+    kv['image'] = 'todo://todo' # TODO: some image path?
+    #kv['image'] = 'https://archive.org/todo' # TODO: some image path?
 
     # TODO
-    dpi = 300
+    dpi = int(page_metadata['estimated_dpi'])
+    scaler = lambda x: int(x * page_metadata['estimated_scale'])
+    #dpi = 300
     kv['scan_res'] = '%d %d' % (dpi, dpi)
 
     pageelem = etree.Element('div', attrib={'class': 'ocr_page',
@@ -274,24 +289,28 @@ def pdf_page_to_hocr_page(page, pageno=None):
         })
 
     for block in pagedata['blocks']:
+        # TODO: Look into negative bbox values
+
         if block['type'] != 0:
             # TODO: Skip blocks that are not text, for now
             continue
 
         kv = {}
-        kv['bbox'] = [str(int(x)) for x in block['bbox']]
+        kv['bbox'] = get_bbox(block['bbox'], scaler)
         blockelem = etree.Element('div', attrib={'class': 'ocr_carea'})
         blockelem.attrib['title'] = assemble_hocr_title_element(kv)
         blockelem.attrib['id'] = get_id(pageno, 'block')
 
-        pdf_process_text_block(pageno, block, blockelem)
+        pdf_process_text_block(pageno, block, blockelem, scaler=scaler)
         pageelem.append(blockelem)
 
     return pageelem
 
 
 
-def process_files(filename):
+def process_files(filename, json_metadata_file):
+    metadata = json.load(open(json_metadata_file))
+
     print('''<?xml version="1.0" encoding="UTF-8"?>
 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
 <html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
@@ -306,7 +325,9 @@ def process_files(filename):
 
     doc = fitz.open(filename)
     for idx, page in enumerate(doc):
-        hocr_page = pdf_page_to_hocr_page(page, pageno=idx)
+        page_metadata = metadata['page_data'][idx]
+
+        hocr_page = pdf_page_to_hocr_page(page, page_metadata, pageno=idx)
         s = etree.tostring(hocr_page, pretty_print=True, method='xml',
                            encoding='utf-8').decode('utf-8')
         print(s)
@@ -319,6 +340,8 @@ if __name__ == '__main__':
     parser = argparse.ArgumentParser(description='Page to hOCR converter')
     parser.add_argument('-f', '--infile', help='Input file',
                         type=str, default=None)
+    parser.add_argument('-J', '--json-metadata-file', help='Input json metadata file',
+                        type=str, default=None)
     args = parser.parse_args()
 
-    process_files(args.infile)
+    process_files(args.infile, args.json_metadata_file)