from __future__ import annotations

import numpy as np

from .chroma import BackgroundModel
from .deps import require_cv2
from .settings import DespillSettings


def _weighted_blur(values: np.ndarray, weights: np.ndarray, radius: int) -> np.ndarray:
    if radius <= 0:
        return values
    cv2 = require_cv2()
    kernel_size = radius * 2 + 1
    numerator = cv2.GaussianBlur(
        values * weights[..., None], (kernel_size, kernel_size), 0
    )
    denominator = cv2.GaussianBlur(weights, (kernel_size, kernel_size), 0)
    return numerator / np.maximum(denominator[..., None], 1e-6)


def despill_green(
    rgb: np.ndarray,
    alpha: np.ndarray,
    bg_confidence: np.ndarray,
    model: BackgroundModel,
    settings: DespillSettings,
) -> tuple[np.ndarray, np.ndarray]:
    rgb_f = rgb.astype(np.float32) / 255.0
    if not settings.enabled:
        return rgb.copy(), np.zeros(alpha.shape, dtype=np.float32)

    cv2 = require_cv2()
    edge = (alpha > settings.edge_low) & (alpha < settings.edge_high)
    if settings.edge_expand_radius > 0:
        radius = settings.edge_expand_radius
        kernel = cv2.getStructuringElement(
            cv2.MORPH_ELLIPSE, (radius * 2 + 1, radius * 2 + 1)
        )
        edge = cv2.dilate(edge.astype(np.uint8), kernel).astype(bool)

    r = rgb_f[..., 0]
    g = rgb_f[..., 1]
    b = rgb_f[..., 2]
    neutral_green = np.maximum(r, b) + settings.green_excess_margin
    excess = np.maximum(g - neutral_green, 0.0)
    alpha_weight = np.maximum(
        np.clip((1.0 - np.abs(alpha - 0.5) * 2.0), 0.0, 1.0),
        settings.alpha_weight_floor,
    )
    bg_weight = (
        settings.bg_confidence_weight * np.clip(bg_confidence, 0.0, 1.0)
        + (1.0 - settings.bg_confidence_weight)
    )
    green_weight = np.clip(excess / max(settings.green_excess_margin * 8.0, 1e-6), 0.0, 1.0)
    mask = edge.astype(np.float32) * alpha_weight * bg_weight * green_weight

    out = rgb_f.copy()
    target_green = np.minimum(g, neutral_green)
    out[..., 1] = g * (1.0 - mask * settings.strength) + target_green * (
        mask * settings.strength
    )

    # Light foreground color approximation: green spill usually removed too much
    # luminance from thin hair edges, so redistribute part of the removed green
    # into red/blue without changing fully opaque pixels.
    bg_green = float(model.rgb_center[1])
    compensation = excess * mask * settings.strength * min(0.25, bg_green * 0.15)
    out[..., 0] = np.clip(out[..., 0] + compensation * 0.5, 0.0, 1.0)
    out[..., 2] = np.clip(out[..., 2] + compensation * 0.5, 0.0, 1.0)

    # Approximate foreground recovery for semi-transparent contaminated pixels.
    # C = alpha * F + (1 - alpha) * B, so F = (C - (1 - alpha) * B) / alpha.
    # Blend it only where green excess is present; this is intentionally local
    # because incorrect unmixing on opaque clothing would damage normal colors.
    bg = np.asarray(model.rgb_center, dtype=np.float32)
    a = np.clip(alpha[..., None], settings.unmix_alpha_min, 1.0)
    recovered = np.clip((rgb_f - (1.0 - alpha[..., None]) * bg) / a, 0.0, 1.0)
    unmix_mask = (
        mask
        * settings.unmix_strength
        * np.clip((1.0 - alpha) / max(1.0 - settings.unmix_alpha_min, 1e-6), 0.0, 1.0)
    )
    out = out * (1.0 - unmix_mask[..., None]) + recovered * unmix_mask[..., None]

    local_fg_weights = (
        (alpha >= settings.local_fg_alpha_threshold)
        & (bg_confidence <= settings.local_fg_bg_confidence_max)
    ).astype(np.float32)
    local_fg = _weighted_blur(out, local_fg_weights, settings.local_fg_blur_radius)
    local_mask = (
        mask
        * settings.local_fg_strength
        * np.clip((1.0 - alpha) / max(1.0 - settings.unmix_alpha_min, 1e-6), 0.0, 1.0)
    )
    out = out * (1.0 - local_mask[..., None]) + local_fg * local_mask[..., None]

    return np.clip(out * 255.0, 0, 255).astype(np.uint8), mask.astype(np.float32)