Spaces:
Runtime error
Runtime error
Update utils/image_utils.py
Browse files- utils/image_utils.py +52 -17
utils/image_utils.py
CHANGED
|
@@ -1,23 +1,58 @@
|
|
| 1 |
-
|
| 2 |
-
import
|
| 3 |
-
from
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
|
| 5 |
def split_image(image: Image.Image, num_chunks: int) -> list:
|
|
|
|
|
|
|
|
|
|
| 6 |
width, height = image.size
|
| 7 |
-
|
|
|
|
|
|
|
| 8 |
chunks = []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
for i in range(num_chunks):
|
| 10 |
-
top = i
|
| 11 |
-
bottom =
|
| 12 |
-
|
| 13 |
-
|
| 14 |
|
| 15 |
-
|
| 16 |
-
buffered = BytesIO()
|
| 17 |
-
image.save(buffered, format="PNG")
|
| 18 |
-
encoded = base64.b64encode(buffered.getvalue()).decode()
|
| 19 |
-
return f"""
|
| 20 |
-
<div style="height:500px; overflow-y:auto; border:1px solid #ccc;">
|
| 21 |
-
<img src="data:image/png;base64,{encoded}" style="width:100%;" />
|
| 22 |
-
</div>
|
| 23 |
-
"""
|
|
|
|
| 1 |
+
import numpy as np
|
| 2 |
+
import cv2
|
| 3 |
+
from PIL import Image
|
| 4 |
+
|
| 5 |
+
def find_low_complexity_row(gray, target_row, search_pct=0.2):
|
| 6 |
+
"""
|
| 7 |
+
Find a nearby row (within ±20%) that has low edge/text density.
|
| 8 |
+
"""
|
| 9 |
+
h, w = gray.shape
|
| 10 |
+
search_radius = int(h * search_pct)
|
| 11 |
+
|
| 12 |
+
start = max(0, target_row - search_radius)
|
| 13 |
+
end = min(h - 1, target_row + search_radius)
|
| 14 |
+
|
| 15 |
+
# Compute edge map
|
| 16 |
+
edges = cv2.Canny(gray, 80, 160)
|
| 17 |
+
|
| 18 |
+
# Sum edges along each row → 1D profile
|
| 19 |
+
row_scores = edges[start:end].sum(axis=1)
|
| 20 |
+
|
| 21 |
+
# Find the row index with the *minimum* score
|
| 22 |
+
best_local_idx = np.argmin(row_scores)
|
| 23 |
+
|
| 24 |
+
# Convert local idx → global row idx
|
| 25 |
+
best_row = start + best_local_idx
|
| 26 |
+
return best_row
|
| 27 |
+
|
| 28 |
|
| 29 |
def split_image(image: Image.Image, num_chunks: int) -> list:
|
| 30 |
+
"""
|
| 31 |
+
Intelligent split: moves split boundaries up/down by ±20% to avoid text.
|
| 32 |
+
"""
|
| 33 |
width, height = image.size
|
| 34 |
+
img_cv = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
|
| 35 |
+
gray = cv2.cvtColor(img_cv, cv2.COLOR_BGR2GRAY)
|
| 36 |
+
|
| 37 |
chunks = []
|
| 38 |
+
split_points = []
|
| 39 |
+
|
| 40 |
+
# First compute approximate "uniform" split points
|
| 41 |
+
approx_points = [int(i * height / num_chunks) for i in range(1, num_chunks)]
|
| 42 |
+
|
| 43 |
+
# Adjust each boundary to a low-density horizontal strip
|
| 44 |
+
for pt in approx_points:
|
| 45 |
+
best = find_low_complexity_row(gray, target_row=pt)
|
| 46 |
+
split_points.append(best)
|
| 47 |
+
|
| 48 |
+
# Add start and end
|
| 49 |
+
split_points = [0] + split_points + [height]
|
| 50 |
+
|
| 51 |
+
# Produce chunks
|
| 52 |
for i in range(num_chunks):
|
| 53 |
+
top = split_points[i]
|
| 54 |
+
bottom = split_points[i+1]
|
| 55 |
+
chunk = image.crop((0, top, width, bottom))
|
| 56 |
+
chunks.append(chunk)
|
| 57 |
|
| 58 |
+
return chunks
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|