File size: 2,588 Bytes
c7498ef | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 | import numpy as np
import cv2
try:
import mediapipe as mp
_HAS_MP = True
except Exception:
_HAS_MP = False
def segment_image(img_np):
"""
Returns a binary mask (uint8 0/255) for hair/head region using MediaPipe if available,
otherwise falls back to naive ellipse.
"""
if not _HAS_MP:
# Fallback to naive ellipse
h, w = img_np.shape[:2]
mask = np.zeros((h, w), dtype=np.uint8)
center = (w // 2, int(h * 0.38))
axes = (int(w * 0.28), int(h * 0.33))
cv2.ellipse(mask, center, axes, 0, 0, 360, 255, -1)
return mask
# Use MediaPipe FaceMesh for better head segmentation
mp_face = mp.solutions.face_mesh
with mp_face.FaceMesh(static_image_mode=True, max_num_faces=1) as face_mesh:
results = face_mesh.process(cv2.cvtColor(img_np, cv2.COLOR_BGR2RGB))
if not results.multi_face_landmarks:
return np.zeros(img_np.shape[:2], dtype=np.uint8) # Empty mask if no face
landmarks = results.multi_face_landmarks[0]
h, w = img_np.shape[:2]
mask = np.zeros((h, w), dtype=np.uint8)
# Define head contour points (approximate hair region using face landmarks)
head_points = [
10, # Top forehead
109, 67, 103, 54, 21, 162, 127, 234, 93, 132, 215, # Left side
338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288, # Right side
152 # Chin
]
points = np.array([(int(landmarks.landmark[i].x * w), int(landmarks.landmark[i].y * h)) for i in head_points])
# Expand slightly for hair
hull = cv2.convexHull(points)
cv2.fillConvexPoly(mask, hull, 255)
# Dilate to cover more hair area
kernel = np.ones((15, 15), np.uint8)
mask = cv2.dilate(mask, kernel, iterations=2)
return mask
def estimate_landmarks(img_np):
"""
Return dict with key landmarks for alignment (e.g., forehead anchor).
"""
if not _HAS_MP:
return None
mp_face = mp.solutions.face_mesh
with mp_face.FaceMesh(static_image_mode=True, max_num_faces=1) as face_mesh:
results = face_mesh.process(cv2.cvtColor(img_np, cv2.COLOR_BGR2RGB))
if not results.multi_face_landmarks:
return None
lm = results.multi_face_landmarks[0].landmark
# Forehead anchor (average of top landmarks)
xs = [p.x for p in lm[:10]] # Normalized [0,1]
ys = [p.y for p in lm[:10]]
h, w = img_np.shape[:2]
return {"forehead_anchor": (int(np.mean(xs) * w), int(np.mean(ys) * h))} |