1. New: Backup the resulting file if it already exists;

2. New: Add `threshold` parameter to control the proportion of face area;
3. Fix: `image_type` did not work;
4. Fix: Empty `log` folder would be created when run `stone -h`;

setup.py

@@ -1,7 +1,7 @@
 from setuptools import setup
 from setuptools import find_packages
 
-VERSION = '0.2.2'
+VERSION = '1.0.0'
 
 with open('README.md') as f:
     LONG_DESCRIPTION = f.read()

src/stone/__main__.py

@@ -29,7 +29,7 @@ def process_image(filename, image_type_setting,
                   specified_palette, default_palette,
                   specified_tone_labels, default_tone_labels,
                   to_bw, new_width, n_dominant_colors,
-                  scale, min_nbrs, min_size, verbose):
+                  scale, min_nbrs, min_size, threshold, verbose):
     basename, extension = filename.stem, filename.suffix
 
     image: np.ndarray = cv2.imread(str(filename.resolve()), cv2.IMREAD_COLOR)
@@ -45,6 +45,8 @@ def process_image(filename, image_type_setting,
     image_type = image_type_setting
     if image_type == 'auto':
         image_type = 'bw' if is_bw else 'color'
+    else:
+        is_bw = image_type == 'bw'
     if len(specified_palette) == 0:
         skin_tone_palette = default_palette['bw' if to_bw or is_bw else 'color']
     else:
@@ -57,7 +59,8 @@ def process_image(filename, image_type_setting,
     try:
         records, report_images = process(image, is_bw, to_bw, skin_tone_palette, tone_labels,
                                          new_width=new_width, n_dominant_colors=n_dominant_colors,
-                                         scaleFactor=scale, minNeighbors=min_nbrs, minSize=min_size, verbose=verbose)
+                                         scaleFactor=scale, minNeighbors=min_nbrs, minSize=min_size, threshold=threshold,
+                                         verbose=verbose)
         return {
             'basename': basename,
             'extension': extension,
@@ -76,6 +79,7 @@ def process_image(filename, image_type_setting,
 
 
 def main():
+    args = build_arguments()
     # Setup logger
     now = datetime.now()
     os.makedirs('./log', exist_ok=True)
@@ -87,8 +91,6 @@ def main():
         datefmt='%H:%M:%S'
     )
 
-    args = build_arguments()
-
     filenames = build_filenames(args.images)
     is_single_file = len(filenames) == 1
 
@@ -127,6 +129,7 @@ def main():
     os.makedirs(output_dir, exist_ok=True)
     result_filename = os.path.join(output_dir, './result.csv')
     image_type_setting = args.image_type
+    threshold = args.threshold
 
     def write_to_csv(row: list):
         with lock:
@@ -139,7 +142,7 @@ def write_to_csv(row: list):
 
     # Backup result.csv if exists
     if os.path.exists(result_filename):
-        renamed_file = os.path.join(output_dir, './result_bak.csv')
+        renamed_file = os.path.join(output_dir, now.strftime('./result_bak_%y%m%d%H%M.csv'))
         shutil.move(result_filename, renamed_file)
     header = 'file,image type,face id,' + ','.join(
         [f'dominant {i + 1},props {i + 1}' for i in range(n_dominant_colors)]) + ',skin tone,PERLA,accuracy(0-100)'
@@ -150,7 +153,7 @@ def write_to_csv(row: list):
                                     specified_palette=specified_palette, default_palette=default_tone_palette,
                                     specified_tone_labels=specified_tone_labels, default_tone_labels=default_tone_labels,
                                     to_bw=to_bw, new_width=new_width, n_dominant_colors=n_dominant_colors,
-                                    scale=scale, min_nbrs=min_nbrs, min_size=min_size, verbose=debug)
+                                    scale=scale, min_nbrs=min_nbrs, min_size=min_size, threshold=threshold, verbose=debug)
 
     with logging_redirect_tqdm():
         with tqdm(filenames, desc='Processing images', unit='images') as pbar:

src/stone/image.py

@@ -56,7 +56,7 @@ def resize(image, width: int = -1, height: int = -1):
     return cv2.resize(image, (width, height))
 
 
-def detect_faces(image, scaleFactor=1.1, minNeighbors=5, minSize=(30, 30), biggest_only=True):
+def detect_faces(image, scaleFactor=1.1, minNeighbors=5, minSize=(30, 30), biggest_only=True, is_bw=False, threshold=0.3):
     gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
     gray = cv2.equalizeHist(gray)
 
@@ -74,7 +74,18 @@ def detect_faces(image, scaleFactor=1.1, minNeighbors=5, minSize=(30, 30), bigge
         return []
     # Change the format of faces from (x, y, w, h) to (x, y, x+w, y+h)
     faces[:, 2:] += faces[:, :2]
-    return faces
+    return [face for face in faces if is_face(face, image, is_bw, threshold)]
+
+
+def is_face(face_coord, image, is_bw, threshold=0.3):
+    x1, y1, x2, y2 = face_coord
+    face_image = image[y1:y2, x1:x2]
+    detect_skin_fn = detect_skin_in_bw if is_bw else detect_skin_in_color
+    _, skin_mask = detect_skin_fn(face_image)
+    skin_pixels = cv2.countNonZero(skin_mask)
+    total_pixels = face_image.shape[0] * face_image.shape[1]
+    skin_ratio = skin_pixels / total_pixels
+    return skin_ratio >= threshold
 
 
 def mask_face(image, face):
@@ -287,12 +298,11 @@ def create_message_bar(dmnt_colors, dmnt_props, tone_hex, distance, bar_width):
 
 
 def process(image: np.ndarray, is_bw: bool, to_bw: bool, skin_tone_palette: list, tone_labels: list = None, new_width=-1, n_dominant_colors=2,
-            scaleFactor=1.1, minNeighbors=5, minSize=(30, 30), biggest_only=True,
-            verbose=False):
+            scaleFactor=1.1, minNeighbors=5, minSize=(30, 30), biggest_only=True, threshold=0.3, verbose=False):
     image = resize(image, new_width)
 
     records, report_images = {}, {}
-    face_coords = detect_faces(image, scaleFactor, minNeighbors, minSize, biggest_only)
+    face_coords = detect_faces(image, scaleFactor, minNeighbors, minSize, biggest_only, is_bw, threshold)
     n_faces = len(face_coords)
 
     if n_faces == 0:

src/stone/utils.py

@@ -88,5 +88,7 @@ def build_arguments():
                              'Higher value results in less detections but with higher quality, defaults to 5', default=5)
     parser.add_argument('--min_size', type=int, nargs='+', metavar=('WIDTH', 'HEIGHT'),
                         help='CONFIG: minimum possible face size. Faces smaller than that are ignored, defaults to "90 90".', default=(90, 90))
+    parser.add_argument('--threshold', type=float, metavar='THRESHOLD', help='CONFIG: what proportion of the skin area is required to identify the '
+                                                                             'face, defaults to 0.3', default=0.3)
 
     return parser.parse_args()