feat: process images (#11)

Adds ability to process images as single page documents. Uses mostly the same path as processing a pdf, but uses a null layout (since an image won't provide a pdf layout) and does not filter layout text blocks using a bounding box to discover text (because there are no layout text blocks).

Author: qued

CHANGELOG.md

@@ -1,5 +1,6 @@
-## 0.2.2-dev0
+## 0.2.2-dev1
 
+* Add capability to process image files
 * Add logic to use OCR when layout text is full of unknown characters
 
 ## 0.2.1

sample-docs/loremipsum.jpg

@@ -178,24 +178,37 @@ def test_process_file_with_model_raises_on_invalid_model_name():
         layout.process_file_with_model("", model_name="fake")
 
 
-class MockPageLayout(layout.PageLayout):
-    def __init__(self, ocr_text):
-        self.ocr_text = ocr_text
-
-    def ocr(self, text_block):
-        return self.ocr_text
+class MockPoints:
+    def tolist(self):
+        return [1, 2, 3, 4]
 
 
 class MockTextBlock(lp.TextBlock):
-    def __init__(self, text):
+    def __init__(self, type=None, text=None, ocr_text=None):
+        self.type = type
         self.text = text
+        self.ocr_text = ocr_text
+
+    @property
+    def points(self):
+        return MockPoints()
+
+
+class MockPageLayout(layout.PageLayout):
+    def __init__(self, layout=None, model=None):
+        self.image = None
+        self.layout = layout
+        self.model = model
+
+    def ocr(self, text_block: MockTextBlock):
+        return text_block.ocr_text
 
 
 def test_interpret_text_block_use_ocr_when_text_symbols_cid():
     fake_text = "(cid:1)(cid:2)(cid:3)(cid:4)(cid:5)"
     fake_ocr = "ocrme"
-    fake_text_block = MockTextBlock(fake_text)
-    assert MockPageLayout(fake_ocr).interpret_text_block(fake_text_block) == fake_ocr
+    fake_text_block = MockTextBlock(text=fake_text, ocr_text=fake_ocr)
+    assert MockPageLayout().interpret_text_block(fake_text_block) == fake_ocr
 
 
 @pytest.mark.parametrize(
@@ -212,3 +225,45 @@ def test_cid_ratio(text, expected):
 )
 def test_is_cid_present(text, expected):
     assert layout.is_cid_present(text) == expected
+
+
+class MockLayout:
+    def __init__(self, *elements):
+        self.elements = elements
+
+    def sort(self, key, inplace):
+        return self.elements
+
+    def __iter__(self):
+        return iter(self.elements)
+
+    def get_texts(self):
+        return [el.text for el in self.elements]
+
+
+def test_pagelayout_without_layout():
+    mock_layout = MockLayout(
+        MockTextBlock(text=None, ocr_text="textblock1"),
+        MockTextBlock(text=None, ocr_text="textblock2"),
+    )
+
+    model = MockLayoutModel(mock_layout)
+    pl = MockPageLayout(model=model, layout=None)
+
+    assert [el.text for el in pl.get_elements(inplace=False)] == [
+        el.ocr_text for el in model.detect(None)
+    ]
+
+
+@pytest.mark.parametrize("filetype", ("png", "jpg"))
+def test_from_image_file(monkeypatch, mock_page_layout, filetype):
+    def mock_get_elements(self, *args, **kwargs):
+        self.elements = [mock_page_layout]
+
+    monkeypatch.setattr(layout.PageLayout, "get_elements", mock_get_elements)
+    elements = (
+        layout.DocumentLayout.from_image_file(f"sample-docs/loremipsum.{filetype}")
+        .pages[0]
+        .elements
+    )
+    assert elements[0] == mock_page_layout

sample-docs/loremipsum.png

@@ -1 +1 @@
-__version__ = "0.2.2-dev0"  # pragma: no cover
+__version__ = "0.2.2-dev1"  # pragma: no cover

test_unstructured_inference/inference/test_layout.py

@@ -72,6 +72,15 @@ def from_file(cls, filename: str, model: Optional[Detectron2LayoutModel] = None)
             pages.append(page)
         return cls.from_pages(pages)
 
+    @classmethod
+    def from_image_file(cls, filename: str, model: Optional[Detectron2LayoutModel] = None):
+        """Creates a DocumentLayout from an image file."""
+        logger.info(f"Reading image file: {filename} ...")
+        image = Image.open(filename)
+        page = PageLayout(number=0, image=image, layout=None, model=model)
+        page.get_elements()
+        return cls.from_pages([page])
+
 
 class PageLayout:
     """Class for an individual PDF page."""
@@ -80,7 +89,7 @@ def __init__(
         self,
         number: int,
         image: Image,
-        layout: lp.Layout,
+        layout: Optional[lp.Layout],
         model: Optional[Detectron2LayoutModel] = None,
     ):
         self.image = image
@@ -107,12 +116,14 @@ def get_elements(self, inplace=True) -> Optional[List[LayoutElement]]:
         # sophisticated ordering logic for more complicated layouts.
         image_layout.sort(key=lambda element: element.coordinates[1], inplace=True)
         for item in image_layout:
-            text_blocks = self.layout.filter_by(item, center=True)
             text = str()
-            for text_block in text_blocks:
-                text_block.text = self.interpret_text_block(text_block)
-            text = " ".join([x for x in text_blocks.get_texts() if x])
-
+            if self.layout is None:
+                text = self.interpret_text_block(item)
+            else:
+                text_blocks = self.layout.filter_by(item, center=True)
+                for text_block in text_blocks:
+                    text_block.text = self.interpret_text_block(text_block)
+                text = " ".join([x for x in text_blocks.get_texts() if x])
             elements.append(
                 LayoutElement(type=item.type, text=text, coordinates=item.points.tolist())
             )