Use normalized glcm and add intensity image

Author: MaximLippeveld

src/scip/features/__init__.py

@@ -80,7 +80,8 @@ def features_partition(
             if "shape" in types:
                 out[i, c:c + lengths["shape"]] = shape_features(
                     mask=p["mask"],
-                    combined_mask=p["combined_mask"]
+                    combined_mask=p["combined_mask"],
+                    image=p["pixels"]
                 )
                 c += lengths["shape"]
             if "raw" in types:

src/scip/features/shape.py

@@ -112,15 +112,21 @@ def _shape_features_meta(channel_names: List[str]) -> Mapping[str, type]:
     return out
 
 
-def _row(mask: numpy.ndarray) -> numpy.ndarray:
+def _row(mask: numpy.ndarray, image: numpy.ndarray = None) -> numpy.ndarray:
     label_img = label(mask)
-    props = regionprops_table(label_image=label_img, properties=prop_ids, cache=False)
+    props = regionprops_table(
+        label_image=label_img,
+        properties=prop_ids,
+        intensity_image=image,
+        cache=False
+    )
     return [numpy.mean(props[k]) for k in prop_names]
 
 
 def shape_features(
     mask: numpy.ndarray,
-    combined_mask: numpy.ndarray
+    combined_mask: numpy.ndarray,
+    image: numpy.ndarray
 ) -> numpy.ndarray:
     """Extracts shape features from image.
 
@@ -139,7 +145,7 @@ def shape_features(
 
     for i in range(len(mask)):
         if numpy.any(mask[i]):
-            out[(i + 1) * len(prop_names):(i + 2) * len(prop_names)] = _row(mask[i])
+            out[(i + 1) * len(prop_names):(i + 2) * len(prop_names)] = _row(mask[i], image[i])
         else:
             # setting proper default values if possible when the mask is empty
             out[(i + 1) * len(prop_names):(i + 2) * len(prop_names)] = [

src/scip/features/texture.py

@@ -64,7 +64,8 @@ def _row(pixels, num_features):
         distances=distances,
         angles=angles,
         levels=bins + 2,
-        symmetric=True
+        symmetric=True,
+        normed=True
     )[:-1, :-1]
 
     out = numpy.full(shape=(num_features,), fill_value=None, dtype=float)