Update detect.py

detect.py

@@ -1,14 +1,45 @@
 import logging
-import random
+
+import cv2
+import numpy as np
 
 logger = logging.getLogger(__name__)
 
+
+def _expand_box(x, y, w, h, img_w, img_h, scale=1.2):
+    pad_w = int(w * (scale - 1) / 2)
+    pad_h = int(h * (scale - 1) / 2)
+    x1 = max(0, x - pad_w)
+    y1 = max(0, y - pad_h)
+    x2 = min(img_w, x + w + pad_w)
+    y2 = min(img_h, y + h + pad_h)
+    return x1, y1, x2, y2
+
+
+def _crop_to_face(img):
+    cv_img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+    gray = cv2.cvtColor(cv_img, cv2.COLOR_BGR2GRAY)
+
+    face_cascade_path = cv2.data.haarcascades + "haarcascade_frontalface_alt2.xml"
+    face_cascade = cv2.CascadeClassifier(face_cascade_path)
+    if face_cascade.empty():
+        face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")
+
+    faces = face_cascade.detectMultiScale(gray, 1.3, 5)
+    if len(faces) == 0:
+        logger.info("No face detected; using full image for classification.")
+        return img
+
+    x, y, w, h = max(faces, key=lambda f: f[2] * f[3])
+    img_h, img_w = gray.shape[:2]
+    x1, y1, x2, y2 = _expand_box(x, y, w, h, img_w, img_h, scale=1.25)
+    return img.crop((x1, y1, x2, y2))
+
 def detect_face_shape(image_path):
     """
     Detects face shape using PIL and trained Mean Face model.
     """
     from PIL import Image
-    import numpy as np
     from classifier import classify_face_shape
 
     try:
@@ -22,6 +53,11 @@ def detect_face_shape(image_path):
         if img is None:
             raise ValueError("Could not load image")
 
+        # Crop to detected face region to improve accuracy
+        if isinstance(image_path, str):
+            img = img.convert("RGB")
+            img = _crop_to_face(img)
+
         # Classify directly (classifier handles resizing/grayscale)
         shape_probs = classify_face_shape(img)