I have a Brother printer. Each time I scan a document, it produces this black and whitish pattern, that looks like a barcode.
However, this is really annoying if while compiling your Italian tax declaration you’d scan 60 documents and each of them has a very long area with this pattern.
I’ve asked to LLM to produce a script that given a reference pattern, it deletes for each image and clean the images.
First, we need to isolate the pattern, therefore we make a copy of any picture, isolate the pattern and save it with a name like pattern.png
Then, we install the required dependencies:
sudo apt update
sudo apt install python3 python3-opencv
We then save into a file named subtract_pattern.py this script:
#!/usr/bin/env python3
from pathlib import Path
import cv2
import numpy as np
import sys
EXTS = {".jpg", ".jpeg", ".png", ".tif", ".tiff", ".bmp"}
def resize_to(img, shape):
return cv2.resize(img, (shape[1], shape[0]), interpolation=cv2.INTER_AREA)
def find_crop_from_pattern(img, pattern, diff_thr=18, min_run=8):
if img.shape != pattern.shape:
pattern = resize_to(pattern, img.shape)
diff = cv2.absdiff(img, pattern)
diff = cv2.GaussianBlur(diff, (5, 5), 0)
# Collassa in valori per riga/colonna
row_score = diff.mean(axis=1)
col_score = diff.mean(axis=0)
# Cerca la zona "diversa dal pattern"
row_mask = row_score > diff_thr
col_mask = col_score > diff_thr
def interval_from_mask(mask):
idx = np.where(mask)[0]
if len(idx) == 0:
return None
# unisci regioni vicine
runs = []
s = idx[0]
prev = idx[0]
for i in idx[1:]:
if i - prev > 1:
runs.append((s, prev))
s = i
prev = i
runs.append((s, prev))
# scegli intervallo principale
runs = [r for r in runs if (r[1] - r[0] + 1) >= min_run]
if not runs:
return None
return max(runs, key=lambda t: t[1] - t[0])
y_int = interval_from_mask(row_mask)
x_int = interval_from_mask(col_mask)
if y_int is None or x_int is None:
return None
y1, y2 = y_int[0], y_int[1] + 1
x1, x2 = x_int[0], x_int[1] + 1
return x1, y1, x2, y2
def main():
if len(sys.argv) < 3:
print("Uso: python3 crop_by_pattern.py <pattern_ref> <input_dir>")
sys.exit(1)
pattern_path = Path(sys.argv[1])
input_dir = Path(sys.argv[2])
out_dir = Path("output_crop")
out_dir.mkdir(exist_ok=True)
pattern = cv2.imread(str(pattern_path), cv2.IMREAD_GRAYSCALE)
if pattern is None:
print("Impossibile leggere il pattern.")
sys.exit(1)
files = sorted([p for p in input_dir.iterdir() if p.suffix.lower() in EXTS])
if not files:
print("Nessuna immagine trovata.")
sys.exit(1)
for p in files:
img = cv2.imread(str(p), cv2.IMREAD_GRAYSCALE)
if img is None:
continue
bbox = find_crop_from_pattern(img, pattern)
if bbox is None:
print(f"{p.name}: crop non trovato, salto")
continue
x1, y1, x2, y2 = bbox
cropped = img[y1:y2, x1:x2]
cv2.imwrite(str(out_dir / p.name), cropped)
print(f"{p.name}: crop {x1},{y1} -> {x2},{y2}")
if __name__ == "__main__":
main()
Finally we can run it:
python3 subtract_pattern.py pattern.png .
# assuming that script is in the same directory and pattern is named pattern.png
And… voilà! In the output_crop folder we have all of our documents without the ugly area!~
Here’s what a document looks like:
