added updates to fix issue #340

pdf2docx/image/ImagesExtractor.py

@@ -296,7 +296,31 @@ def _to_raw_dict(image: fitz.Pixmap, bbox: fitz.Rect):
         Returns:
             dict: Raw dict of the pixmap.
         """
-        if image.colorspace.n > 3:  # must convert: we only support PNG
+        """
+        if image.colorspace.n > 3: # must convert: we only support PNG image with alpha channel
+            image = fitz.Pixmap(fitz.csRGB, image)
+        """
+        # added by maksym
+        # Get the colorspace object of the pixmap.
+        # NOTE: colorspace.n (number of channels) is NOT sufficient to decide
+        # whether MuPDF can write this pixmap as PNG.
+        cs = image.colorspace
+        # MuPDF can write PNG only if the colorspace is exactly:
+        #   - DeviceGray (fitz.csGRAY)
+        #   - DeviceRGB  (fitz.csRGB)
+        #
+        # Even if cs.n == 1 or cs.n == 3, the pixmap may still be:
+        #   - ICCBased grayscale/RGB
+        #   - Separation / spot color
+        #   - Indexed colorspace
+        #   - Alpha-only pixmap
+        #
+        # All of these will trigger:
+        #   "pixmap must be grayscale or rgb to write as png"
+        #
+        # Therefore we must convert unless the colorspace object itself
+        # is exactly csGRAY or csRGB.
+        if cs is None or cs not in (fitz.csGRAY, fitz.csRGB):
             image = fitz.Pixmap(fitz.csRGB, image)
         return {
             "type": BlockType.IMAGE.value,

test/test.py

@@ -287,6 +287,25 @@ def test_rotated_images(self):
         docx_file = os.path.join(output_path, f'{filename}.docx')
         assert os.path.isfile(docx_file), f'Expected output file: {docx_file}'
         assert os.path.getsize(docx_file) > 0, 'Output docx should not be empty'
+
+    # ------------------------------------------
+    # non-grayscale/non-RGB images (issue 340)
+    # ------------------------------------------
+    def test_non_grayscale_non_rgb_images(self):
+        '''Test converting PDFs with rotated images and non-RGB colorspace (CMYK, alpha).
+
+        Covers: per-image rotation, pixmap tobytes() for non-grayscale/non-RGB.
+        '''
+        filenames = [
+            'demo-issue-340'
+        ]
+        for filename in filenames:
+            pdf_file = os.path.join(sample_path, f'{filename}.pdf')
+            if not os.path.isfile(pdf_file):
+                continue
+            docx_file = os.path.join(output_path, f'{filename}.docx')
+            parse(pdf_file, docx_file, start=0, end=None)
+            assert os.path.isfile(docx_file), f'Expected output {docx_file}'
 
 
 # We make a separate pytest test for each sample file.