"""Image quality metrics — brightness, contrast, sharpness, noise, composite.

Consolidates the heuristic quality scoring that was previously duplicated
between the image_quality provider and the duplicate_detector provider.
"""

from __future__ import annotations

import cv2
import numpy as np

from cores.vision.color import to_gray


def brightness(img: np.ndarray) -> float:
    """Mean pixel intensity (0-255)."""
    return float(np.mean(to_gray(img)))


def contrast(img: np.ndarray) -> float:
    """Standard deviation of pixel intensity."""
    return float(np.std(to_gray(img)))


def sharpness(img: np.ndarray) -> float:
    """Variance of Laplacian — higher = sharper."""
    gray = to_gray(img)
    return float(cv2.Laplacian(gray, cv2.CV_64F).var())


def noise_level(img: np.ndarray) -> float:
    """Estimate noise via median absolute deviation of the Laplacian.

    Robust, simple, no model required.
    """
    gray = to_gray(img)
    lap = cv2.Laplacian(gray, cv2.CV_64F)
    return float(np.median(np.abs(lap - np.median(lap))) / 0.6745)


def quality_score(img: np.ndarray) -> float:
    """Composite 0-1 quality score (heuristic).

    Combines brightness, contrast, sharpness, and noise into a single
    0-1 score where 1.0 = excellent quality.
    """
    b = brightness(img)
    c = contrast(img)
    s = sharpness(img)
    n = noise_level(img)

    # Brightness: ideal ~128
    b_score = 1.0 - min(1.0, abs(b - 128.0) / 128.0)
    # Contrast: ideal std ~50-80
    c_score = 1.0 - min(1.0, abs(c - 60.0) / 100.0)
    # Sharpness: log-scale, ~100 = good, ~1000 = excellent
    s_score = min(1.0, np.log1p(s) / np.log1p(1000.0))
    # Noise: lower is better; >20 is bad
    n_score = max(0.0, 1.0 - n / 30.0)
    return 0.25 * (b_score + c_score + s_score + n_score)