recolor/main.py

import argparse
import sys
from pathlib import Path
import multiprocessing
from itertools import product

import numpy as np
from rich.console import Console
from scipy.stats import moment
from skimage.io import imread, imsave
from PIL import Image
from numba import njit, prange
from scipy.spatial import KDTree

console = Console()


def save_image(data, name, resolution):
    final_image = data.reshape(resolution)
    imsave(name, final_image)


def find_nearest_point(data, target):
    idx = np.array([calc_distance(p, target) for p in data]).argmin()
    return data[idx]


def centroidnp(arr):
    length = arr.shape[0]
    sum_x = np.sum(arr[:, 0])
    sum_y = np.sum(arr[:, 1])
    sum_z = np.sum(arr[:, 2])
    return sum_x / length, sum_y / length, sum_z / length


def calc_distance(x, y):
    return np.absolute(np.linalg.norm(x - y))


def k_means(data, count):
    # Pick n random points to startA
    idx_data = np.unique(data, axis=0)
    index = np.random.choice(idx_data.shape[0], count, replace=False)
    means = idx_data[index]
    data = np.delete(data, index, axis=0)

    distance_delta = 100
    means_distance = 0
    clusters = {}
    with console.status("[bold blue] Finding means...") as status:
        while distance_delta > 5:
            # Initialize cluster map
            clusters = {}
            for m in means:
                clusters[repr(m)] = []

            # Find closest mean to each point
            for point in data:
                closest = find_nearest_point(means, point)
                clusters[repr(closest)].append(point)

            # Find the centroid of each mean
            new_means = []
            previous_distance = means_distance
            means_distance = 0
            for mean in means:
                mean_key = repr(mean)
                # Clean up the results a little bit
                clusters[mean_key] = np.stack(clusters[mean_key])
                # Calculate new mean
                raw_mean = centroidnp(clusters[mean_key])
                nearest_mean_point = find_nearest_point(data, raw_mean)
                means_distance = means_distance + calc_distance(
                    mean, nearest_mean_point
                )
                new_means.append(nearest_mean_point)
            means_distance = means_distance / float(count)
            distance_delta = abs(previous_distance - means_distance)
            means = np.stack(new_means)

        return means


def find_closest_points(points1, points2):
    # Build a k-d tree from the points in the second array
    tree = KDTree(points2)

    # Find the closest point in the second array for each element in the first array
    closest_points_indices = tree.query(points1)[1]
    closest_points = points2[closest_points_indices]

    # Return the result
    return closest_points


def main(image_name, output_name, colors):
    im = imread(image_name)
    starting_resolution = im.shape
    im_resized = np.array(Image.fromarray(im).resize(size=(150, 150)))
    console.log("[blue] Starting with image of size: ", starting_resolution)
    raw_pixels_resized = im_resized.reshape(-1, 3)

    # Find the colors that most represent the image
    color_means = k_means(raw_pixels_resized, len(colors))
    console.log("[green] Found cluster centers: ", color_means)

    # Remap image to the center points
    raw_pixels = im.reshape(-1, 3)
    raw_shape = raw_pixels.shape
    console.log("[purple] Re-mapping image")
    output_raw = np.zeros_like(raw_pixels)
    output_raw = find_closest_points(raw_pixels, color_means)
    console.log("[purple] Re-mapping image complete (phase 1)")


    # Map means to the colors provided by the user
    pairs = []
    tmp_means = color_means
    for color in colors:
        m = find_nearest_point(tmp_means, color)
        pairs.append((m, color))
        (idxs,) = np.where(np.all(tmp_means == m, axis=1))
        tmp_means = np.delete(tmp_means, idxs, axis=0)

    # Recolor the image
    for pair in pairs:
        (idxs,) = np.where(np.all(output_raw == pair[0], axis=1))
        output_raw[idxs] = pair[1]

    if output_name is None:
        output_name = f"{input_image.parents[0]}/recolored-{input_image.name}"
    save_image(output_raw, output_name, starting_resolution)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(
        prog="Recolor",
        description="Recolor changes the color palette of an image to the one provided by the user",
    )
    parser.add_argument(
        "filename", help="Input image"
    )
    parser.add_argument(
        "-o", "--output", help="Output image destination, defaults to the same path as the input titled [path]/recolored-[name]", dest="outname"
    )
    color_loader_group = parser.add_mutually_exclusive_group(required=True)
    color_loader_group.add_argument(
        "-f",
        "--file",
        help="A file of RGB color values, with one color per line",
        dest="cfpath",
    )
    color_loader_group.add_argument(
        "-l",
        "--list",
        nargs='+',
        help="A list of RGB color values. Example: '123,90,89 212,7,0'",
        dest="clist",
    )
    # Check inputs
    args = parser.parse_args()
    input_image = Path(args.filename)
    if not input_image.exists():
        console.log("[red] Error: input image path does not exist")
        sys.exit(-1)
    else:  
        if not input_image.is_file():
            console.log("[red] Error: input image path is not a file")
            sys.exit(-1)

    # Parse out color
    colors = []

    if args.cfpath is not None:
        # Load from file
        lines = []
        input_colors = Path(args.cfpath)
        with open(input_colors) as f:
            lines = [line.rstrip() for line in f]
        # Split each line
        for line in lines:
            vals = []
            sp = line.split(',')
            if len(sp) != 3:
                console.log("[red] Error: RGB value in file is malformed -- not exactly 3 values", args.cfpath, line)
                sys.exit(-2)
            for v in sp:
                vals.append(int(v))
            colors.append(np.array(vals))
    else:
        #Load from list
        for line in args.clist:
            vals = []
            sp = line.split(',')
            if len(sp) != 3:
                console.log("[red] Error: RGB value in list is malformed -- not exactly 3 values", line)
                sys.exit(-2)
            for v in sp:
                vals.append(int(v))
            colors.append(np.array(vals))

    main(input_image, Path(args.outname) if args.outname is not None else None, np.array(colors))