Updating verification engines to include latest updates to redactor engines (#1162)

* Enabling use of ad-hoc recognizers in verifier

* Adding support to standard image verification engine as well

* Linting fix

* Removing redundant init

* Removing unused import

Author: niwilso

presidio-image-redactor/presidio_image_redactor/dicom_image_pii_verify_engine.py

@@ -49,15 +49,21 @@ def verify_dicom_instance(
         instance: pydicom.dataset.FileDataset,
         padding_width: int = 25,
         display_image: bool = True,
+        use_metadata: bool = True,
         ocr_kwargs: Optional[dict] = None,
+        ad_hoc_recognizers: Optional[List[PatternRecognizer]] = None,
         **text_analyzer_kwargs,
     ) -> Tuple[Optional[PIL.Image.Image], dict, list]:
         """Verify PII on a single DICOM instance.
 
         :param instance: Loaded DICOM instance including pixel data and metadata.
         :param padding_width: Padding width to use when running OCR.
         :param display_image: If the verificationimage is displayed and returned.
+        :param use_metadata: Whether to redact text in the image that
+        are present in the metadata.
         :param ocr_kwargs: Additional params for OCR methods.
+        :param ad_hoc_recognizers: List of PatternRecognizer objects to use
+        for ad-hoc recognizer.
         :param text_analyzer_kwargs: Additional values for the analyze method
         in ImageAnalyzerEngine.
 
@@ -82,24 +88,17 @@ def verify_dicom_instance(
             loaded_image = Image.open(png_filepath)
             image = self._add_padding(loaded_image, is_greyscale, padding_width)
 
-        # Create custom recognizer using DICOM metadata
-        original_metadata, is_name, is_patient = self._get_text_metadata(instance_copy)
-        phi_list = self._make_phi_list(original_metadata, is_name, is_patient)
-        deny_list_recognizer = PatternRecognizer(
-            supported_entity="PERSON", deny_list=phi_list
-        )
+        # Get analyzer results
         ocr_results = self.ocr_engine.perform_ocr(image)
-        analyzer_results = self.image_analyzer_engine.analyze(
-            image,
-            ocr_kwargs=ocr_kwargs,
-            ad_hoc_recognizers=[deny_list_recognizer],
-            **text_analyzer_kwargs,
+        analyzer_results = self._get_analyzer_results(
+            image, instance, use_metadata, ocr_kwargs, ad_hoc_recognizers,
+            **text_analyzer_kwargs
         )
 
         # Get image with verification boxes
         verify_image = (
             self.verify(
-                image, ad_hoc_recognizers=[deny_list_recognizer], **text_analyzer_kwargs
+                image, ad_hoc_recognizers=ad_hoc_recognizers, **text_analyzer_kwargs
             )
             if display_image
             else None
@@ -114,7 +113,9 @@ def eval_dicom_instance(
         padding_width: int = 25,
         tolerance: int = 50,
         display_image: bool = False,
+        use_metadata: bool = True,
         ocr_kwargs: Optional[dict] = None,
+        ad_hoc_recognizers: Optional[List[PatternRecognizer]] = None,
         **text_analyzer_kwargs,
     ) -> Tuple[Optional[PIL.Image.Image], dict]:
         """Evaluate performance for a single DICOM instance.
@@ -124,7 +125,11 @@ def eval_dicom_instance(
         :param padding_width: Padding width to use when running OCR.
         :param tolerance: Pixel distance tolerance for matching to ground truth.
         :param display_image: If the verificationimage is displayed and returned.
+        :param use_metadata: Whether to redact text in the image that
+        are present in the metadata.
         :param ocr_kwargs: Additional params for OCR methods.
+        :param ad_hoc_recognizers: List of PatternRecognizer objects to use
+        for ad-hoc recognizer.
         :param text_analyzer_kwargs: Additional values for the analyze method
         in ImageAnalyzerEngine.
 
@@ -135,7 +140,9 @@ def eval_dicom_instance(
             instance,
             padding_width,
             display_image,
+            use_metadata,
             ocr_kwargs=ocr_kwargs,
+            ad_hoc_recognizers=ad_hoc_recognizers,
             **text_analyzer_kwargs,
         )
         formatted_ocr_results = self.bbox_processor.get_bboxes_from_ocr_results(

presidio-image-redactor/presidio_image_redactor/image_pii_verify_engine.py

@@ -1,9 +1,10 @@
 from PIL import Image, ImageChops
-from presidio_image_redactor.image_analyzer_engine import ImageAnalyzerEngine
+from presidio_image_redactor.image_redactor_engine import ImageRedactorEngine
+from presidio_analyzer import PatternRecognizer
 import matplotlib
 import io
 from matplotlib import pyplot as plt
-from typing import Optional
+from typing import Optional, List
 
 
 def fig2img(fig):
@@ -16,16 +17,15 @@ def fig2img(fig):
     return img
 
 
-class ImagePiiVerifyEngine:
+class ImagePiiVerifyEngine(ImageRedactorEngine):
     """ImagePiiVerifyEngine class only supporting Pii verification currently."""
 
-    def __init__(self, image_analyzer_engine: Optional[ImageAnalyzerEngine] = None):
-        if not image_analyzer_engine:
-            image_analyzer_engine = ImageAnalyzerEngine()
-        self.image_analyzer_engine = image_analyzer_engine
-
     def verify(
-        self, image: Image, ocr_kwargs: Optional[dict] = None, **text_analyzer_kwargs
+        self,
+        image: Image,
+        ocr_kwargs: Optional[dict] = None,
+        ad_hoc_recognizers: Optional[List[PatternRecognizer]] = None,
+        **text_analyzer_kwargs
     ) -> Image:
         """Annotate image with the detect PII entity.
 
@@ -34,6 +34,8 @@ def verify(
 
         :param image: PIL Image to be processed.
         :param ocr_kwargs: Additional params for OCR methods.
+        :param ad_hoc_recognizers: List of PatternRecognizer objects to use
+        for ad-hoc recognizer.
         :param text_analyzer_kwargs: Additional values for the analyze method
         in ImageAnalyzerEngine.
 
@@ -42,9 +44,23 @@ def verify(
 
         image = ImageChops.duplicate(image)
         image_x, image_y = image.size
-        bboxes = self.image_analyzer_engine.analyze(
-            image, ocr_kwargs, **text_analyzer_kwargs
-        )
+
+        # Detect PII
+        self._check_ad_hoc_recognizer_list(ad_hoc_recognizers)
+        if ad_hoc_recognizers is None:
+            bboxes = self.image_analyzer_engine.analyze(
+                image,
+                ocr_kwargs=ocr_kwargs,
+                **text_analyzer_kwargs,
+            )
+        else:
+            bboxes = self.image_analyzer_engine.analyze(
+                image,
+                ocr_kwargs=ocr_kwargs,
+                ad_hoc_recognizers=ad_hoc_recognizers,
+                **text_analyzer_kwargs,
+            )
+
         fig, ax = plt.subplots()
         image_r = 70
         fig.set_size_inches(image_x / image_r, image_y / image_r)

presidio-image-redactor/tests/test_dicom_image_pii_verify_engine.py

@@ -86,21 +86,11 @@ def test_verify_dicom_instance_happy_path(
     mock_add_padding = mocker.patch.object(
         DicomImagePiiVerifyEngine, "_add_padding", return_value=None
     )
-    mock_get_metadata = mocker.patch.object(
-        DicomImagePiiVerifyEngine, "_get_text_metadata", return_value=[None, None, None]
-    )
-    mock_make_phi_list = mocker.patch.object(
-        DicomImagePiiVerifyEngine, "_make_phi_list", return_value=None
-    )
-    mock_patternrecognizer = mocker.patch(
-        "presidio_image_redactor.dicom_image_pii_verify_engine.PatternRecognizer",
-        return_value=None,
-    )
     mock_perform_ocr = mocker.patch.object(
         TesseractOCR, "perform_ocr", return_value=None
     )
     mock_analyze = mocker.patch.object(
-        ImageAnalyzerEngine, "analyze", return_value=None
+        DicomImagePiiVerifyEngine, "_get_analyzer_results", return_value=None
     )
     mock_verify = mocker.patch.object(
         DicomImagePiiVerifyEngine, "verify", return_value=None
@@ -115,9 +105,6 @@ def test_verify_dicom_instance_happy_path(
     assert mock_save_pixel_array.call_count == 1
     assert mock_image_open.call_count == 1
     assert mock_add_padding.call_count == 1
-    assert mock_get_metadata.call_count == 1
-    assert mock_make_phi_list.call_count == 1
-    assert mock_patternrecognizer.call_count == 1
     assert mock_perform_ocr.call_count == 1
     assert mock_analyze.call_count == 1
     assert mock_verify.call_count == 1