import cv2
import numpy as np
from PIL import Image
import mediapipe as mp

mp_face_mesh = mp.solutions.face_mesh

def preprocess_image(image_path: str) -> np.ndarray:
    """Crop face region using MediaPipe, return normalized array for ONNX."""
    img = cv2.imread(image_path)
    img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

    with mp_face_mesh.FaceMesh(static_image_mode=True, max_num_faces=1) as face_mesh:
        results = face_mesh.process(img_rgb)
        if results.multi_face_landmarks:
            h, w = img.shape[:2]
            landmarks = results.multi_face_landmarks[0].landmark
            xs = [int(l.x * w) for l in landmarks]
            ys = [int(l.y * h) for l in landmarks]
            x1, x2 = max(min(xs)-20, 0), min(max(xs)+20, w)
            y1, y2 = max(min(ys)-20, 0), min(max(ys)+20, h)
            face_crop = img_rgb[y1:y2, x1:x2]
        else:
            face_crop = img_rgb  # fallback: use full image

    # Resize and normalize for MobileNetV2 / ResNet50
    resized = cv2.resize(face_crop, (224, 224))
    arr = resized.astype(np.float32) / 255.0
    arr = (arr - [0.485, 0.456, 0.406]) / [0.229, 0.224, 0.225]  # ImageNet norm
    return arr.transpose(2, 0, 1)[np.newaxis, :]  # (1, 3, 224, 224)


def delete_image(image_path: str):
    """DPDP compliance: delete raw image after inference."""
    import os
    if os.path.exists(image_path):
        os.remove(image_path)