Worked on feature extraction

.vscode/settings.json

@@ -9,5 +9,12 @@
     "python.testing.pytestEnabled": false,
     "python.testing.nosetestsEnabled": false,
     "python.testing.unittestEnabled": true,
-    "python.pythonPath": "/home/lukas/.local/share/virtualenvs/shenzhen-solitaire-nsu5dgrx/bin/python"
+    "python.linting.mypyArgs": [
+        "--ignore-missing-imports",
+        "--follow-imports=silent",
+        "--show-column-numbers",
+        "--strict"
+    ],
+    "python.linting.mypyEnabled": true,
+    "python.formatting.provider": "black"
 }

tools/feature_extraction.py

@@ -4,10 +4,10 @@ from shenzhen_solitaire.board import NumberCard, SpecialCard
 import cv2
 
 import numpy as np
-from typing import Any, Tuple
+from typing import Any, Tuple, List, Union, Dict, Optional
 
 
-def border_image(image, size=1, color=0):
+def border_image(image: np.array, size: int = 1, color: int = 0) -> None:
     for ring in range(size):
         for x in range(ring, image.shape[0] - ring):
             image[x][ring] = color
@@ -17,47 +17,39 @@ def border_image(image, size=1, color=0):
             image[image.shape[0] - 1 - ring][y] = color
 
 
-def prepare_image(image):
+def prepare_image(image: np.array) -> np.array:
     cnt = get_contour(image)
     mask = np.zeros(image.shape[:2], dtype=image.dtype)
     contim = cv2.drawContours(mask, [cnt], 0, 255, cv2.FILLED)
-    # crop = np.multiply(edge_image, contim)
     return contim
 
 
-def get_contour(image):
+def get_contour(image: np.array) -> np.array:
     gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
     ret, edge_image = cv2.threshold(gray_image, 127, 255, cv2.THRESH_BINARY_INV)
-    border_image(edge_image, size=0)
+    kernel = cv2.getStructuringElement(cv2.MORPH_CROSS, (3, 3))
+    edge_image = cv2.morphologyEx(edge_image, cv2.MORPH_CLOSE, kernel)
+    border_image(edge_image, size=1)
     contours, hierarchy = cv2.findContours(
         edge_image, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE
     )
     cnt = max(contours, key=cv2.contourArea)
+    assert isinstance(cnt, np.ndarray)
     return cnt
 
 
-def matchScaleInvShape(cont1, cont2):
+def matchScaleInvShape(cont1: np.array, cont2: np.array) -> float:
     m1 = cv2.moments(cont1)
     m2 = cv2.moments(cont2)
     moments = [
         (m1[moment], m2[moment]) for moment in m1 if str(moment).startswith("nu")
     ]
-    return sum([abs((nu1) - (nu2)) for nu1, nu2 in moments])
+    return sum([abs((nu1) - (nu2)) * 1000 for nu1, nu2 in moments])
 
 
-def match_template(image, template):
+def type_fine(
-    image_cont, hierarchy = cv2.findContours(
+    one: Union[SpecialCard, NumberCard], other: Union[SpecialCard, NumberCard]
-        image, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE
+) -> bool:
-    )
-    imcont = max(image_cont, key=cv2.contourArea)
-    template_cont, hierarchy = cv2.findContours(
-        template, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE
-    )
-    temcont = max(template_cont, key=cv2.contourArea)
-    return [matchScaleInvShape(imcont, temcont)]
-
-
-def type_fine(one, other) -> bool:
     if isinstance(one, SpecialCard):
         return one == other
     assert isinstance(one, NumberCard)
@@ -65,7 +57,10 @@ def type_fine(one, other) -> bool:
         return False
     return one.number == other.number
 
-def check_type(matches, should_type):
+
+def check_type(
+    matches: List[Any], should_type: Union[SpecialCard, NumberCard]
+) -> Optional[int]:
     if not type_fine(matches[0][0], should_type):
         correct_index = 0
         for list_type, list_value, _ in matches:
@@ -77,24 +72,43 @@ def check_type(matches, should_type):
             f"{str(should_type):>20} matched as {str(matches[0][0]):>20} {matches[0][1]:.05f}, "
             f"correct in pos {correct_index:02d} val {correct_value:.05f}"
         )
-        cv2.imshow("one", prepare_image(catalogue[matches[0][2]][0]))
+        catalogue_index = matches[correct_index][2]
-        cv2.imshow("two", img1)
+        assert isinstance(catalogue_index, int)
-        cv2.imshow("three", prepare_image(catalogue[matches[correct_index][2]][0]))
+        return catalogue_index
-        cv2.waitKey(0)
+    return None
-        return True
-    return False
 
-def debug_match(image, image_type, catalogue):
+
+def show_wrong_images(
+    current: np.ndarray, correct: np.ndarray, wrong: np.ndarray
+) -> None:
+    cv2.imshow("Current", current)
+    cv2.imshow("Correct", correct)
+    cv2.imshow("Wrong", wrong)
+    cv2.waitKey(0)
+
+
+def debug_match(
+    image: np.array,
+    image_type: Union[NumberCard, SpecialCard],
+    catalogue: List[Tuple[Any, Union[SpecialCard, NumberCard]]],
+) -> None:
     cnt1 = prepare_image(image)
     i1_matches = []
     for index, (template_image, template_type) in enumerate(catalogue):
         cnt2 = prepare_image(template_image)
         i1_matches.append((template_type, matchScaleInvShape(cnt1, cnt2), index))
     i1_matches = sorted(i1_matches, key=lambda x: x[1])
-    
+    correct_type_index = check_type(i1_matches, image_type)
+    if correct_type_index is not None:
+        show_wrong_images(
+            cnt1,
+            prepare_image(catalogue[correct_type_index][0]),
+            prepare_image(catalogue[i1_matches[0][2]][0]),
+        )
+        return
     for list_type, list_value, list_index in i1_matches:
         if not type_fine(list_type, i1_matches[0][0]):
-            if list_value * 0.4 < i1_matches[0][1]:
+            if list_value * 0.8 < i1_matches[0][1]:
                 print(
                     f"{str(image_type):>20} {i1_matches[0][1]:.05f} very close"
                     f" match with {str(list_type):>20} {list_value:.05f}"