| import cv2 |
| import numpy as np |
| import mediapipe as mp |
| from typing import List, Tuple, Optional |
|
|
| class FaceDetector: |
| def __init__(self, min_detection_confidence=0.5): |
| self.mp_face_detection = mp.solutions.face_detection |
| self.mp_face_mesh = mp.solutions.face_mesh |
| self.face_detection = self.mp_face_detection.FaceDetection( |
| model_selection=1, |
| min_detection_confidence=min_detection_confidence |
| ) |
| self.face_mesh = self.mp_face_mesh.FaceMesh( |
| max_num_faces=1, |
| min_detection_confidence=min_detection_confidence, |
| min_tracking_confidence=min_detection_confidence |
| ) |
| |
| def detect_faces(self, image: np.ndarray) -> List[dict]: |
| """ |
| Detect faces in image |
| """ |
| rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) |
| results = self.face_detection.process(rgb_image) |
| |
| faces = [] |
| if results.detections: |
| for detection in results.detections: |
| bbox = detection.location_data.relative_bounding_box |
| faces.append({ |
| 'bbox': { |
| 'x': bbox.xmin, |
| 'y': bbox.ymin, |
| 'width': bbox.width, |
| 'height': bbox.height |
| }, |
| 'confidence': detection.score[0] |
| }) |
| |
| return faces |
| |
| def get_face_landmarks(self, image: np.ndarray) -> Optional[List[Tuple[float, float]]]: |
| """ |
| Get face landmarks for detailed analysis |
| """ |
| rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) |
| results = self.face_mesh.process(rgb_image) |
| |
| if results.multi_face_landmarks: |
| landmarks = [] |
| h, w, _ = image.shape |
| for landmark in results.multi_face_landmarks[0].landmark: |
| landmarks.append((landmark.x * w, landmark.y * h)) |
| return landmarks |
| |
| return None |
| |
| def get_face_regions(self, image: np.ndarray) -> dict: |
| """ |
| Extract different face regions for analysis |
| """ |
| landmarks = self.get_face_landmarks(image) |
| if not landmarks: |
| return {} |
| |
| h, w, _ = image.shape |
| |
| |
| regions = { |
| 'forehead': (int(w * 0.3), int(h * 0.1), int(w * 0.4), int(h * 0.2)), |
| 'left_cheek': (int(w * 0.1), int(h * 0.3), int(w * 0.2), int(h * 0.3)), |
| 'right_cheek': (int(w * 0.7), int(h * 0.3), int(w * 0.2), int(h * 0.3)), |
| 'nose': (int(w * 0.4), int(h * 0.3), int(w * 0.2), int(h * 0.2)), |
| 'chin': (int(w * 0.4), int(h * 0.6), int(w * 0.2), int(h * 0.2)), |
| } |
| |
| return regions |
| |
| def crop_face(self, image: np.ndarray, padding: float = 0.2) -> Optional[np.ndarray]: |
| """ |
| Crop face from image with padding |
| """ |
| faces = self.detect_faces(image) |
| if not faces: |
| return None |
| |
| face = faces[0] |
| h, w, _ = image.shape |
| |
| x = int(face['bbox']['x'] * w) |
| y = int(face['bbox']['y'] * h) |
| width = int(face['bbox']['width'] * w) |
| height = int(face['bbox']['height'] * h) |
| |
| |
| pad_x = int(width * padding) |
| pad_y = int(height * padding) |
| |
| x1 = max(0, x - pad_x) |
| y1 = max(0, y - pad_y) |
| x2 = min(w, x + width + pad_x) |
| y2 = min(h, y + height + pad_y) |
| |
| return image[y1:y2, x1:x2] |
| |
| def get_face_angle(self, image: np.ndarray) -> float: |
| """ |
| Estimate face angle (for pose detection) |
| """ |
| landmarks = self.get_face_landmarks(image) |
| if not landmarks: |
| return 0.0 |
| |
| |
| left_eye = landmarks[33] |
| right_eye = landmarks[263] |
| |
| dx = right_eye[0] - left_eye[0] |
| dy = right_eye[1] - left_eye[1] |
| |
| angle = np.arctan2(dy, dx) * 180 / np.pi |
| return angle |
