def test_depth(pool):
"using greater depth should trace out the mask edge more closely"
image = np.zeros((1000, 1000))
rr, cc = draw.circle(300, 500, 150)
image[rr, cc] = 1
image = pm.rescale_commensurate(image, (5, 5), depth=2)
mask = image.astype(bool)
tiles0 = pm.partition(image, (5, 5), mask=mask, depth=0)
tiles1 = pm.partition(image, (5, 5), mask=mask, depth=1)
tiles2 = pm.partition(image, (5, 5), mask=mask, depth=2)
assert len(tiles0) < len(tiles1) < len(tiles2)
def test_depth(pool):
"using greater depth should trace out the mask edge more closely"
image = np.zeros((1000, 1000))
rr, cc = draw.circle(300, 500, 150)
image[rr, cc] = 1
image = pm.rescale_commensurate(image, (5, 5), depth=2)
mask = image.astype(bool)
tiles0 = pm.partition(image, (5, 5), mask=mask, depth=0)
tiles1 = pm.partition(image, (5, 5), mask=mask, depth=1)
tiles2 = pm.partition(image, (5, 5), mask=mask, depth=2)
assert len(tiles0) < len(tiles1) < len(tiles2)
def detail_mosaic(filename, dirname):
from skimage.io import imread
image = imread(filename)
# Size the image to be evenly divisible by the tiles.
from skimage import img_as_float
image = img_as_float(image)
# Use perceptually uniform colorspace for all analysis.
import photomosaic as pm
converted_img = pm.perceptual(image)
pool = pm.make_pool(dirname + '/*.jpg')
# Adapt the color palette of the image to resemble the palette of the pool.
adapted_img = pm.adapt_to_pool(converted_img, pool)
scaled_img = pm.rescale_commensurate(adapted_img,
grid_dims=(30, 30),
depth=1)
tiles = pm.partition(scaled_img, grid_dims=(30, 30), depth=1)
annotated_img = pm.draw_tile_layout(pm.rgb(scaled_img), tiles)
from skimage.io import imsave
imsave(filename[:-4] + '_detail_mosaic' + filename[-4:], annotated_img)
def create_mosaic(input_img, images_dataset_path, grid_dims):
# Load a sample image
image = cv2.imread(img_path)
image = img_as_float(image) #ensure image is float ranging from 0 to 1
# Analyze the collection (the "pool") of images.
pool = pm.make_pool(images_dataset_path)
# Use perceptually uniform colorspace for all analysis.
converted_img = pm.perceptual(image)
# Adapt the color palette of the image to resemble the palette of the pool.
#adapted_img = pm.adapt_to_pool(converted_img, pool)
adapted_img = converted_img
#scale = 1
#scaled_img = Image.new('RGB', (adapted_img.shape[0] * scale, adapted_img.shape[1] * scale))
#scaled_img = Image.new('RGB', (5040, 5040))
scaled_img = pm.rescale_commensurate(adapted_img,
grid_dims=grid_dims,
depth=0)
tiles = pm.partition(scaled_img, grid_dims=grid_dims, depth=0)
# Reshape the 3D array (height, width, color_channels) into
# a 2D array (num_pixels, color_channels) and average over the pixels.
tile_colors = [
np.mean(scaled_img[tile].reshape(-1, 3), 0) for tile in tiles
]
# Match a pool image to each tile.
match = pm.simple_matcher(pool)
matches = [match(tc) for tc in tile_colors]
matches_list = [x[0] for x in matches]
# Perform neural network object detection to see what classes are on which images
detect_images(matches_list)
# Concatenate list of matches images to a single mosaic image
concatenateImages(matches_list, grid_dims)
import os
import numpy as np
import photomosaic as pm
from skimage.io import imsave
from skimage.data import chelsea
from skimage import img_as_float
here = os.path.dirname(__file__)
POOL_PATH = '/tmp/photomosaic-docs-pool/pool.json'
pool = pm.import_pool(os.path.join(POOL_PATH))
image = img_as_float(chelsea())
converted_img = pm.perceptual(image)
scaled_img = pm.rescale_commensurate(converted_img, grid_dims=(30, 30),
depth=0)
tiles = pm.partition(scaled_img, grid_dims=(30, 30), depth=0)
tile_colors = [np.mean(scaled_img[tile].reshape(-1, 3), 0)
for tile in tiles]
match = pm.simple_matcher(pool)
matches = [match(tc) for tc in tile_colors]
canvas = np.ones_like(scaled_img) # white canvas
mos = pm.draw_mosaic(canvas, tiles, matches)
imsave(os.path.join(here, '..', '_static', 'generated_images',
'no-palette-adjustment.png'), mos)
adapted_img = pm.adapt_to_pool(converted_img, pool)
imsave(os.path.join(here, '..', '_static', 'generated_images',
'adapted-chelsea.png'), pm.rgb(adapted_img))
orig_img.load()
W, H = orig_img.size
Hx = 0
Wx = W
screen = pygame.display.set_mode((W + Wx, H + Hx))
draw(orig_img, (0, 0))
pygame.display.flip()
if tune:
img = pm.tune(
orig_img,
database) # Adjust colors levels to what's availabe in the pool.
else:
img = orig_img
tiles = pm.partition(img, (10, 10), depth=DEPTH)
for tile in sorted(tiles, key=analyze_sort):
pm.analyze_one(tile)
tx, ty = pm.tile_position(tile)
w, h = tile.size
for (x, y), color in zip(locs, tile.rgb):
rect = (Wx + tx + w * x / 2, Hx + ty + h * y / 2, w / 2, h / 2)
pygame.draw.rect(screen, color, rect)
pygame.draw.rect(screen, (0, 0, 0), rect, 1)
pygame.display.flip()
db = pm.connect(database)
try:
pm.reset_usage(db)
for tile in sorted(tiles, key=match_sort):
orig_img = pm.open( infile )
orig_img.load()
W,H = orig_img.size
Hx = 0
Wx = W
screen = pygame.display.set_mode((W+Wx,H+Hx))
draw(orig_img, (0,0))
pygame.display.flip()
if tune:
img = pm.tune(orig_img, database) # Adjust colors levels to what's availabe in the pool.
else:
img = orig_img
tiles = pm.partition(img, (10, 10), depth=DEPTH)
for tile in sorted(tiles, key=analyze_sort):
pm.analyze_one(tile)
tx,ty = pm.tile_position(tile)
w,h = tile.size
for (x,y), color in zip(locs, tile.rgb):
rect = (Wx+tx + w*x/2,Hx+ty+h*y/2,w/2,h/2)
pygame.draw.rect(screen, color, rect)
pygame.draw.rect(screen, (0,0,0), rect, 1)
pygame.display.flip()
db = pm.connect(database)
try:
pm.reset_usage(db)
for tile in sorted(tiles, key=match_sort):