Box Fitting Img

Math essence

Same non-overlapping expansion algorithm as boxFitting, but each box samples its color from a procedural 'sky' substrate image at its birth location. The result is a fitted-box mosaic where the substrate color is preserved only at fitted positions, surrounded by a white border.

Parameters

Reference

First 60 lines of reference.py. Full source: library/boxFittingImg/reference.py

from __future__ import annotations

from dataclasses import dataclass, field
from typing import ClassVar

import imageio.v2 as imageio
import numpy as np


def _sky_substrate(width: int, height: int, seed: int) -> np.ndarray:
    """Procedural 'sky' substrate mimicking the tornado/sky photos Tarbell used."""
    rng = np.random.default_rng(seed + 7919)
    yy = np.linspace(0.0, 1.0, height, dtype=np.float32)[:, None]
    xx = np.linspace(0.0, 1.0, width, dtype=np.float32)[None, :]
    sky_top = np.array([25.0, 35.0, 70.0], dtype=np.float32)
    sky_bot = np.array([180.0, 170.0, 160.0], dtype=np.float32)
    grad = yy
    base = sky_top[None, None, :] * (1.0 - grad[:, :, None]) + sky_bot[None, None, :] * grad[:, :, None]
    radial = np.sqrt((xx - 0.5) ** 2 + (yy - 0.55) ** 2)
    swirl = 0.5 + 0.5 * np.sin(50.0 * radial + 8.0 * np.arctan2(yy - 0.5, xx - 0.5))
    perturb = (swirl[..., None] - 0.5) * np.array([40.0, 25.0, -10.0], dtype=np.float32)[None, None, :]
    noise = rng.normal(0.0, 12.0, size=(height, width, 3)).astype(np.float32)
    img = np.clip(base + perturb + noise, 0, 255).astype(np.uint8)
    return img


@dataclass
class _Box:
    x: int
    y: int
    d: int
    color: np.ndarray
    ok_to_draw: bool = False
    chaste: bool = True


class BoxFittingImg:
    family: ClassVar[str] = "growth_aggregation"
    sketch: ClassVar[str] = "boxFittingImg"

    def __init__(
        self,
        dimx: int = 510,
        dim: int = 330,
        dimborder: int = 10,
        maxnum: int = 2000,
        initial_boxes: int = 5,
        bg_color: tuple[int, int, int] = (0, 0, 0),
        seed: int | None = None,
        substrate_seed: int | None = None,
    ) -> None:
        self.dimx = dimx
        self.dim = dim
        self.dimborder = dimborder
        self.maxnum = maxnum
        self.initial_boxes = initial_boxes
        self.bg_color = np.array(bg_color, dtype=np.uint8)
        self._seed = seed
        self.substrate_seed = substrate_seed
        self.rng = np.random.default_rng(seed)