dotfiles/.config/matugen/scripts/least_busy_region.py
2025-06-20 14:52:11 +02:00

338 lines
14 KiB
Python
Executable file

#!/usr/bin/env python3
# Disclaimer: This script was ai-generated and went through minimal revision.
import os
os.environ["OPENCV_LOG_LEVEL"] = "SILENT"
import cv2
import numpy as np
import argparse
import json
def center_crop(img, target_w, target_h):
h, w = img.shape[:2]
if w == target_w and h == target_h:
return img
x1 = max(0, (w - target_w) // 2)
y1 = max(0, (h - target_h) // 2)
x2 = x1 + target_w
y2 = y1 + target_h
return img[y1:y2, x1:x2]
def find_least_busy_region(image_path, region_width=300, region_height=200, screen_width=None, screen_height=None, verbose=False, stride=2, screen_mode="fill", padding=50):
img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
if img is None:
raise FileNotFoundError(f"Image not found: {image_path}")
orig_h, orig_w = img.shape
scale = 1.0
if screen_width is not None and screen_height is not None:
scale_w = screen_width / orig_w
scale_h = screen_height / orig_h
if screen_mode == "fill":
scale = max(scale_w, scale_h)
else:
scale = min(scale_w, scale_h)
new_w = int(orig_w * scale)
new_h = int(orig_h * scale)
if verbose:
print(f"Scaling image from {orig_w}x{orig_h} to {new_w}x{new_h} (scale: {scale:.3f}, mode: {screen_mode})")
img = cv2.resize(img, (new_w, new_h), interpolation=cv2.INTER_LANCZOS4)
img = center_crop(img, screen_width, screen_height)
if verbose:
print(f"Cropped image to {screen_width}x{screen_height}")
else:
if verbose:
print(f"Using original image size: {orig_w}x{orig_h}")
arr = img.astype(np.float64)
h, w = arr.shape
# Use OpenCV's integral for fast computation
integral = cv2.integral(arr, sdepth=cv2.CV_64F)[1:,1:]
integral_sq = cv2.integral(arr**2, sdepth=cv2.CV_64F)[1:,1:]
def region_sum(ii, x1, y1, x2, y2):
total = ii[y2, x2]
if x1 > 0:
total -= ii[y2, x1-1]
if y1 > 0:
total -= ii[y1-1, x2]
if x1 > 0 and y1 > 0:
total += ii[y1-1, x1-1]
return total
min_var = None
min_coords = (0, 0)
area = region_width * region_height
x_start = padding
y_start = padding
x_end = w - region_width - padding + 1
y_end = h - region_height - padding + 1
for y in range(y_start, max(y_end, y_start+1), stride):
for x in range(x_start, max(x_end, x_start+1), stride):
x1, y1 = x, y
x2, y2 = x + region_width - 1, y + region_height - 1
s = region_sum(integral, x1, y1, x2, y2)
s2 = region_sum(integral_sq, x1, y1, x2, y2)
mean = s / area
var = (s2 / area) - (mean ** 2)
if (min_var is None) or (var < min_var):
min_var = var
min_coords = (x, y)
return min_coords, min_var
def find_largest_region(image_path, screen_width=None, screen_height=None, verbose=False, stride=2, screen_mode="fill", threshold=100.0, aspect_ratio=1.0, padding=50):
img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
if img is None:
raise FileNotFoundError(f"Image not found: {image_path}")
orig_h, orig_w = img.shape
scale = 1.0
if screen_width is not None and screen_height is not None:
scale_w = screen_width / orig_w
scale_h = screen_height / orig_h
if screen_mode == "fill":
scale = max(scale_w, scale_h)
else:
scale = min(scale_w, scale_h)
new_w = int(orig_w * scale)
new_h = int(orig_h * scale)
if verbose:
print(f"Scaling image from {orig_w}x{orig_h} to {new_w}x{new_h} (scale: {scale:.3f}, mode: {screen_mode})")
img = cv2.resize(img, (new_w, new_h), interpolation=cv2.INTER_LANCZOS4)
img = center_crop(img, screen_width, screen_height)
if verbose:
print(f"Cropped image to {screen_width}x{screen_height}")
else:
if verbose:
print(f"Using original image size: {orig_w}x{orig_h}")
arr = img.astype(np.float64)
h, w = arr.shape
# Use OpenCV's integral for fast computation
integral = cv2.integral(arr, sdepth=cv2.CV_64F)[1:,1:]
integral_sq = cv2.integral(arr**2, sdepth=cv2.CV_64F)[1:,1:]
def region_sum(ii, x1, y1, x2, y2):
total = ii[y2, x2]
if x1 > 0:
total -= ii[y2, x1-1]
if y1 > 0:
total -= ii[y1-1, x2]
if x1 > 0 and y1 > 0:
total += ii[y1-1, x1-1]
return total
min_size = 10
max_size = min(h, int(w / aspect_ratio)) if aspect_ratio >= 1.0 else min(int(h * aspect_ratio), w)
best = None
best_size = min_size
while min_size <= max_size:
mid = (min_size + max_size) // 2
if aspect_ratio >= 1.0:
region_h = mid
region_w = int(mid * aspect_ratio)
else:
region_w = mid
region_h = int(mid / aspect_ratio)
if region_w > w or region_h > h:
max_size = mid - 1
continue
found = False
x_start = padding
y_start = padding
x_end = w - region_w - padding + 1
y_end = h - region_h - padding + 1
for y in range(y_start, max(y_end, y_start+1), stride):
for x in range(x_start, max(x_end, x_start+1), stride):
x1, y1 = x, y
x2, y2 = x + region_w - 1, y + region_h - 1
s = region_sum(integral, x1, y1, x2, y2)
s2 = region_sum(integral_sq, x1, y1, x2, y2)
area = region_w * region_h
mean = s / area
var = (s2 / area) - (mean ** 2)
if var <= threshold:
found = True
best = (x, y, region_w, region_h, var)
break
if found:
break
if found:
best_size = mid
min_size = mid + 1
else:
max_size = mid - 1
if best:
x, y, region_w, region_h, var = best
center_x = x + region_w // 2
center_y = y + region_h // 2
return (center_x, center_y), (region_w, region_h), var
else:
return None, (0, 0), None
def draw_region(image_path, coords, region_width=300, region_height=200, output_path='output.png', screen_width=None, screen_height=None, screen_mode="fill"):
img = cv2.imread(image_path)
if img is None:
raise FileNotFoundError(f"Image not found: {image_path}")
orig_h, orig_w = img.shape[:2]
if screen_width is not None and screen_height is not None:
scale_w = screen_width / orig_w
scale_h = screen_height / orig_h
if screen_mode == "fill":
scale = max(scale_w, scale_h)
else:
scale = min(scale_w, scale_h)
new_w = int(orig_w * scale)
new_h = int(orig_h * scale)
img = cv2.resize(img, (new_w, new_h), interpolation=cv2.INTER_LANCZOS4)
img = center_crop(img, screen_width, screen_height)
x, y = coords
cv2.rectangle(img, (x, y), (x+region_width-1, y+region_height-1), (0,0,255), 3)
cv2.imwrite(output_path, img)
print(f"Saved output image with rectangle at {output_path}")
def draw_largest_region(image_path, center, size, output_path='output.png', screen_width=None, screen_height=None, screen_mode="fill"):
img = cv2.imread(image_path)
if img is None:
raise FileNotFoundError(f"Image not found: {image_path}")
orig_h, orig_w = img.shape[:2]
if screen_width is not None and screen_height is not None:
scale_w = screen_width / orig_w
scale_h = screen_height / orig_h
if screen_mode == "fill":
scale = max(scale_w, scale_h)
else:
scale = min(scale_w, scale_h)
new_w = int(orig_w * scale)
new_h = int(orig_h * scale)
img = cv2.resize(img, (new_w, new_h), interpolation=cv2.INTER_LANCZOS4)
img = center_crop(img, screen_width, screen_height)
cx, cy = center
region_w, region_h = size
x1 = cx - region_w // 2
y1 = cy - region_h // 2
x2 = cx + region_w // 2 - 1
y2 = cy + region_h // 2 - 1
cv2.rectangle(img, (x1, y1), (x2, y2), (255,0,0), 3)
cv2.imwrite(output_path, img)
print(f"Saved output image with largest region at {output_path}")
def get_dominant_color(image_path, x, y, w, h, screen_width=None, screen_height=None, screen_mode="fill"):
img = cv2.imread(image_path)
if img is None:
raise FileNotFoundError(f"Image not found: {image_path}")
orig_h, orig_w = img.shape[:2]
if screen_width is not None and screen_height is not None:
scale_w = screen_width / orig_w
scale_h = screen_height / orig_h
if screen_mode == "fill":
scale = max(scale_w, scale_h)
else:
scale = min(scale_w, scale_h)
new_w = int(orig_w * scale)
new_h = int(orig_h * scale)
img = cv2.resize(img, (new_w, new_h), interpolation=cv2.INTER_LANCZOS4)
img = center_crop(img, screen_width, screen_height)
# Ensure region is within bounds
x = max(0, x)
y = max(0, y)
w = max(1, min(w, img.shape[1] - x))
h = max(1, min(h, img.shape[0] - y))
region = img[y:y+h, x:x+w]
if region.size == 0 or region.shape[0] == 0 or region.shape[1] == 0:
return [0, 0, 0]
region = region.reshape((-1, 3))
# Filter out black pixels (optional, improves accuracy for some images)
non_black = region[np.any(region > 10, axis=1)]
if non_black.shape[0] == 0:
non_black = region
region = np.float32(non_black)
if region.shape[0] < 3:
return [int(x) for x in np.mean(region, axis=0)]
# K-means to find dominant color
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 10, 1.0)
K = min(3, region.shape[0])
_, labels, centers = cv2.kmeans(region, K, None, criteria, 10, cv2.KMEANS_RANDOM_CENTERS)
counts = np.bincount(labels.flatten())
dominant = centers[np.argmax(counts)]
return [int(x) for x in dominant]
def main():
parser = argparse.ArgumentParser(description="Find least busy region in an image and output a JSON. Made for determining a suitable position for a wallpaper widget.")
parser.add_argument("image_path", help="Path to the input image")
parser.add_argument("--width", type=int, default=300, help="Region width")
parser.add_argument("--height", type=int, default=200, help="Region height")
parser.add_argument("-v", "--visual-output", action="store_true", help="Output image with rectangle")
parser.add_argument("--screen-width", type=int, default=1920, help="Screen width for wallpaper scaling")
parser.add_argument("--screen-height", type=int, default=1080, help="Screen height for wallpaper scaling")
parser.add_argument("--stride", type=int, default=4, help="Step size for sliding window (higher is faster, less precise)")
parser.add_argument("--screen-mode", choices=["fill", "fit"], default="fill", help="Wallpaper scaling mode: 'fill' (default) or 'fit'")
parser.add_argument("--verbose", action="store_true", help="Print verbose output")
parser.add_argument("-l", "--largest-region", action="store_true", help="Find the largest region under the variance threshold and output its center")
parser.add_argument("-t", "--variance-threshold", type=float, default=1000.0, help="Variance threshold for largest region mode")
parser.add_argument("--aspect-ratio", type=float, default=1.78, help="Aspect ratio (width/height) for largest region mode")
parser.add_argument("--padding", type=int, default=50, help="Minimum distance from region to image edge (default: 50)")
args = parser.parse_args()
if args.largest_region:
center, size, var = find_largest_region(
args.image_path,
screen_width=args.screen_width,
screen_height=args.screen_height,
verbose=args.verbose,
stride=args.stride,
screen_mode=args.screen_mode,
threshold=args.variance_threshold,
aspect_ratio=args.aspect_ratio,
padding=args.padding
)
if center:
if args.visual_output:
draw_largest_region(args.image_path, center, size, screen_width=args.screen_width, screen_height=args.screen_height, screen_mode=args.screen_mode)
# Extract dominant color
cx, cy = center
region_w, region_h = size
x1 = cx - region_w // 2
y1 = cy - region_h // 2
dominant_color = get_dominant_color(
args.image_path, x1, y1, region_w, region_h,
screen_width=args.screen_width, screen_height=args.screen_height, screen_mode=args.screen_mode
)
dominant_color_hex = '#{:02x}{:02x}{:02x}'.format(*dominant_color)
print(json.dumps({
"center_x": center[0],
"center_y": center[1],
"width": size[0],
"height": size[1],
"variance": var,
"dominant_color": dominant_color_hex
}))
else:
print(json.dumps({"error": "No region found under the threshold."}))
return
coords, variance = find_least_busy_region(
args.image_path,
region_width=args.width,
region_height=args.height,
screen_width=args.screen_width,
screen_height=args.screen_height,
verbose=args.verbose,
stride=args.stride,
screen_mode=args.screen_mode,
padding=args.padding
)
if args.visual_output:
draw_region(args.image_path, coords, region_width=args.width, region_height=args.height, screen_width=args.screen_width, screen_height=args.screen_height, screen_mode=args.screen_mode)
# Output JSON with center point
center_x = coords[0] + args.width // 2
center_y = coords[1] + args.height // 2
dominant_color = get_dominant_color(
args.image_path, coords[0], coords[1], args.width, args.height,
screen_width=args.screen_width, screen_height=args.screen_height, screen_mode=args.screen_mode
)
dominant_color_hex = '#{:02x}{:02x}{:02x}'.format(*dominant_color)
print(json.dumps({
"center_x": center_x,
"center_y": center_y,
"width": args.width,
"height": args.height,
"variance": variance,
"dominant_color": dominant_color_hex
}))
if __name__ == "__main__":
main()