added the tesseract_model.py

Signed-off-by: Peter Staar <taa@zurich.ibm.com>

docling/models/tesseract_model.py

@@ -2,6 +2,7 @@ import logging
 from typing import Iterable
 
 import numpy
+import pandas as pd
 
 from docling.datamodel.base_models import BoundingBox, CoordOrigin, OcrCell, Page
 from docling.datamodel.pipeline_options import TesseractOcrOptions
@@ -9,8 +10,8 @@ from docling.models.base_ocr_model import BaseOcrModel
 
 _log = logging.getLogger(__name__)
 
-
 class TesseractModel(BaseOcrModel):
+
     def __init__(self, enabled: bool, options: TesseractOcrOptions):
         super().__init__(enabled=enabled, options=options)
         self.options: TesseractOcrOptions
@@ -18,10 +19,65 @@ class TesseractModel(BaseOcrModel):
         self.scale = 3  # multiplier for 72 dpi == 216 dpi.
 
         if self.enabled:
-            import tesserocr
+            try:
+                self._get_name_and_version()
+                
+            except Exception as exc:
+                _log.error(f"Tesseract is not supported, aborting ...")
+                self.enabled = False
+                
+    def _get_name_and_version(self) -> Tuple[str, str]:
 
-            self.reader = easyocr.Reader(lang_list=self.options.lang)
+        if self._name!=None and self._version!=None:
+            return self._name, self._version
 
+        cmd = ['tesseract', '--version']
+
+        proc = Popen(cmd, stdout=PIPE, stderr=PIPE)
+        stdout, stderr = proc.communicate()
+
+        proc.wait()
+
+        # HACK: Windows versions of Tesseract output the version to stdout, Linux versions 
+        # to stderr, so check both.
+        version_line = (stdout.decode('utf8').strip() or stderr.decode('utf8').strip()).split('\n')[0].strip()
+
+        # If everything else fails...
+        if not version_line:
+            version_line = 'tesseract XXX'
+
+        name, version = version_line.split(' ')
+
+        self._name    = name
+        self._version = version
+
+        return name, version
+
+    def _run_tesseract(self, ifilename, languages=None):
+
+        cmd = ["tesseract"]
+
+        if languages:
+            cmd += ['-l', '+'.join(languages)]
+
+        cmd += [ifilename, 'stdout', "tsv"]
+        logger.info("command: {}".format(" ".join(cmd)))
+
+        proc = Popen(cmd, stdout=PIPE)
+        output, _ = proc.communicate()
+
+        # Read the TSV file generated by Tesseract
+        df = pd.read_csv('output_file_name.tsv', sep='\t')
+
+        # Display the dataframe (optional)
+        print(df.head())
+        
+        # Filter rows that contain actual text (ignore header or empty rows)
+        df_filtered = df[df['text'].notnull() & (df['text'].str.strip() != '')]
+
+        return df_filtered
+        
+    
     def __call__(self, page_batch: Iterable[Page]) -> Iterable[Page]:
 
         if not self.enabled:
@@ -36,37 +92,55 @@ class TesseractModel(BaseOcrModel):
                 high_res_image = page._backend.get_page_image(
                     scale=self.scale, cropbox=ocr_rect
                 )
-                im = numpy.array(high_res_image)
+                print(high_res_image)
-                result = self.reader.readtext(im)
+                
+                # FIXME: do we really need to save the image to a file
+                fname = "temporary-file.png"
+                high_res_image.save(fname)
 
-                del high_res_image
+                if os.path.exists(fname):
-                del im
+                    df = self._run_tesseract(fname)
+                    os.remove(fname)
+                else:
+                    _log.error(f"no image file: {fname}")
+                
+                # Print relevant columns (bounding box and text)
+                for index, row in df_filtered.iterrows():
+                    print(row)
+                    
+                    text = row["text"]
+                    conf = row["confidence"]
+                    
+                    l = float(row['left'])
+                    t = float(row['top'])
+                    w = float(row['width'])
+                    h = float(row['height'])
 
-                cells = [
+                    b = t-h
-                    OcrCell(
+                    r = l+w
+                    
+                    cell = OcrCell(
                         id=ix,
-                        text=line[1],
+                        text=text,
                         confidence=line[2],
                         bbox=BoundingBox.from_tuple(
                             coord=(
-                                (line[0][0][0] / self.scale) + ocr_rect.l,
+                                (l / self.scale) + ocr_rect.l,
-                                (line[0][0][1] / self.scale) + ocr_rect.t,
+                                (b / self.scale) + ocr_rect.t,
-                                (line[0][2][0] / self.scale) + ocr_rect.l,
+                                (r / self.scale) + ocr_rect.l,
-                                (line[0][2][1] / self.scale) + ocr_rect.t,
+                                (t / self.scale) + ocr_rect.t,
                             ),
                             origin=CoordOrigin.TOPLEFT,
                         ),
                     )
-                    for ix, line in enumerate(result)
+                    all_ocr_cells.append(cell)
-                ]
+                
-                all_ocr_cells.extend(cells)
-
             ## Remove OCR cells which overlap with programmatic cells.
             filtered_ocr_cells = self.filter_ocr_cells(all_ocr_cells, page.cells)
 
             page.cells.extend(filtered_ocr_cells)
 
             # DEBUG code:
-            # self.draw_ocr_rects_and_cells(page, ocr_rects)
+            self.draw_ocr_rects_and_cells(page, ocr_rects)
 
             yield page