feat: Fixed char ordering in text lines (#138)

Signed-off-by: Peter Staar <[email protected]>
Signed-off-by: Christoph Auer <[email protected]>
Co-authored-by: Christoph Auer <[email protected]>

Author: PeterStaar-IBM

.github/workflows/wheels.yml

@@ -269,7 +269,11 @@ jobs:
         shell: pwsh
         run: |
           New-Item -Path 'C:\nasm' -ItemType Directory -Force
-          Invoke-WebRequest -Uri 'https://fossies.org/windows/misc/nasm-2.16.03-win64.zip/nasm-2.16.03/nasm.exe' -OutFile 'C:\nasm\nasm.exe'
+          Invoke-WebRequest -Uri 'https://www.nasm.us/pub/nasm/releasebuilds/2.16.03/win64/nasm-2.16.03-win64.zip' -OutFile 'C:\nasm-2.16.03-win64.zip'
+          Expand-Archive -Path 'C:\nasm-2.16.03-win64.zip' -DestinationPath 'C:\nasm-temp' -Force
+          Copy-Item -Path 'C:\nasm-temp\nasm-2.16.03\nasm.exe' -Destination 'C:\nasm\nasm.exe'
+          Remove-Item -Path 'C:\nasm-2.16.03-win64.zip' -Force
+          Remove-Item -Path 'C:\nasm-temp' -Recurse -Force
           nasm -v
 
       - name: Build wheels

docling_parse/pdf_parser.py

@@ -115,7 +115,12 @@ def _to_table_of_contents(self, toc: dict) -> List[PdfTableOfContents]:
         return result
 
     def get_page(
-        self, page_no: int, create_words: bool = True, create_textlines: bool = True
+        self,
+        page_no: int,
+        *,
+        create_words: bool = True,
+        create_textlines: bool = True,
+        enforce_same_font: bool = True,
     ) -> SegmentedPdfPage:
         if page_no in self._pages.keys():
             return self._pages[page_no]
@@ -134,6 +139,7 @@ def get_page(
                         page=page["original"],
                         create_words=create_words,
                         create_textlines=create_textlines,
+                        enforce_same_font=enforce_same_font,
                     )  # put on cache
                     return self._pages[page_no]
 
@@ -315,7 +321,12 @@ def _to_lines(self, data: dict) -> List[PdfLine]:
         return result
 
     def _to_segmented_page(
-        self, page: dict, create_words: bool, create_textlines: bool
+        self,
+        page: dict,
+        *,
+        create_words: bool,
+        create_textlines: bool,
+        enforce_same_font: bool = True,
     ) -> SegmentedPdfPage:
 
         char_cells = self._to_cells(page["cells"])
@@ -330,14 +341,22 @@ def _to_segmented_page(
         )
 
         if create_words:
-            self._create_word_cells(segmented_page)
+            self._create_word_cells(segmented_page, enforce_same_font=enforce_same_font)
 
         if create_textlines:
-            self._create_textline_cells(segmented_page)
+            self._create_textline_cells(
+                segmented_page, enforce_same_font=enforce_same_font
+            )
+
         return segmented_page
 
     def _create_word_cells(
-        self, segmented_page: SegmentedPdfPage, _loglevel: str = "fatal"
+        self,
+        segmented_page: SegmentedPdfPage,
+        *,
+        space_width_factor_for_merge: float = 0.33,
+        enforce_same_font: bool = True,
+        _loglevel: str = "fatal",
     ):
 
         if len(segmented_page.word_cells) > 0:
@@ -355,7 +374,11 @@ def _create_word_cells(
 
         sanitizer.set_char_cells(data=char_data)
 
-        data = sanitizer.create_word_cells(space_width_factor_for_merge=0.33)
+        # data = sanitizer.create_word_cells(space_width_factor_for_merge=0.33)
+        data = sanitizer.create_word_cells(
+            space_width_factor_for_merge=space_width_factor_for_merge,
+            enforce_same_font=enforce_same_font,
+        )
 
         segmented_page.word_cells = []
         for item in data:
@@ -365,9 +388,14 @@ def _create_word_cells(
         segmented_page.has_words = len(segmented_page.word_cells) > 0
 
     def _create_textline_cells(
-        self, segmented_page: SegmentedPdfPage, _loglevel: str = "fatal"
+        self,
+        segmented_page: SegmentedPdfPage,
+        *,
+        space_width_factor_for_merge: float = 1.0,
+        space_width_factor_for_merge_with_space: float = 0.33,
+        enforce_same_font: bool = True,
+        _loglevel: str = "fatal",
     ):
-
         if len(segmented_page.textline_cells) > 0:
             return
 
@@ -387,7 +415,12 @@ def _create_textline_cells(
 
         sanitizer.set_char_cells(data=char_data)
 
-        data = sanitizer.create_line_cells()
+        # data = sanitizer.create_line_cells()
+        data = sanitizer.create_line_cells(
+            space_width_factor_for_merge=space_width_factor_for_merge,
+            space_width_factor_for_merge_with_space=space_width_factor_for_merge_with_space,
+            enforce_same_font=enforce_same_font,
+        )
 
         segmented_page.textline_cells = []
         for item in data:

docling_parse/visualize.py

@@ -76,6 +76,13 @@ def parse_args():
         help="Enable interactive mode (default: False)",
     )
 
+    # Add an optional boolean argument for enforcing same-font
+    parser.add_argument(
+        "--enforce-same-font",
+        action="store_true",
+        help="Enable interactive mode (default: False)",
+    )
+
     # Add an argument for the output directory, defaulting to "./tmp"
     parser.add_argument(
         "-o",
@@ -115,18 +122,21 @@ def parse_args():
         int(args.page),
         args.display_text,
         args.log_text,
+        args.enforce_same_font,
         args.page_boundary,
         args.category,
     )
 
 
 def visualise_py(
+    *,
     log_level: str,
     pdf_path: str,
     interactive: str,
     output_dir: Path,
     display_text: bool,
     log_text: bool,
+    enforce_same_font: bool,
     page_boundary: str = "crop_box",  # media_box
     category: str = "char",  # "both", "sanitized", "original"
     page_num: int = -1,
@@ -142,7 +152,12 @@ def visualise_py(
     for page_no in page_nos:
         print(f"parsing {pdf_path} on page: {page_no}")
 
-        pdf_page: SegmentedPdfPage = pdf_doc.get_page(page_no=page_no)
+        pdf_page: SegmentedPdfPage = pdf_doc.get_page(
+            page_no=page_no,
+            create_words=True,
+            create_textlines=True,
+            enforce_same_font=enforce_same_font,
+        )
 
         if os.path.exists(str(output_dir)):
             pdf_page.save_as_json(
@@ -234,6 +249,7 @@ def main():
         page_num,
         display_text,
         log_text,
+        enforce_same_font,
         page_boundary,
         category,
     ) = parse_args()
@@ -247,6 +263,7 @@ def main():
         output_dir=output_dir,
         display_text=display_text,
         log_text=log_text,
+        enforce_same_font=enforce_same_font,
         page_boundary=page_boundary,
         category=category,
         page_num=page_num,

src/pybind/docling_sanitizer.h

@@ -249,7 +249,7 @@ namespace docling
     // do a deep copy
     word_cells = char_cells;
 
-    LOG_S(INFO) << "#-wordcells: " << word_cells.size();
+    LOG_S(INFO) << "#-word cells: " << word_cells.size();
 
     // remove all spaces 
     auto itr = word_cells.begin();
@@ -265,7 +265,7 @@ namespace docling
 	  }
       }
 
-    LOG_S(INFO) << "#-wordcells: " << word_cells.size();
+    LOG_S(INFO) << "#-word cells: " << word_cells.size();
 
     // > space_width_factor_for_merge, so nothing gets merged with a space
     double space_width_factor_for_merge_with_space = 2.0*space_width_factor_for_merge; 

src/v2/pdf_resources/page_cells.h

@@ -32,6 +32,7 @@ namespace pdflib
     itr_type end() { return cells.end(); }
 
     itr_type erase(itr_type itr) { return cells.erase(itr); }
+    itr_type erase(itr_type itr_0, itr_type itr_1) { return cells.erase(itr_0, itr_1); }
 
     pdf_resource<PAGE_CELL>& at(std::size_t i) { return cells.at(i); }
 

src/v2/pdf_resources/page_font.h

@@ -1555,7 +1555,7 @@ namespace pdflib
     if(cmap_initialized) // we found a `ToUnicode` before. No need to go deeper! 
       {
 	LOG_S(WARNING) << "We found a `ToUnicode` before. No need to go deeper!";
-	// return;
+	return;
       }
     //else