"""Cut the three load-bearing blocks out of the rebuilt page raster and wrap them
in a single self-contained HTML so the page can be read as an image (never as the
OCR text layer).
Blocks come from graham-page.py's position map of journal page 84:
THEOREM ~y 822 px (statement + its displayed formula)
Barban and Vehov ~y 1543 px (display (5))
COROLLARY ~y 1924 px (statement + both cases)
"""
import base64
import io
import sys
from PIL import Image
SRC = sys.argv[1] if len(sys.argv) > 1 else "p84.png"
OUT = sys.argv[2] if len(sys.argv) > 2 else "graham-p84.html"
BLOCKS = [
("THEOREM + main term", 780, 1090),
("display (5): Barban and Vehov", 1500, 1720),
("COROLLARY, both cases", 1880, 2190),
]
def data_uri(img):
buf = io.BytesIO()
img.save(buf, format="PNG")
return "data:image/png;base64," + base64.b64encode(buf.getvalue()).decode()
page = Image.open(SRC).convert("L")
parts = []
for label, y0, y1 in BLOCKS:
crop = page.crop((0, y0, page.width, y1))
crop = crop.resize((int(crop.width * 1.7), int(crop.height * 1.7)), Image.LANCZOS)
parts.append(
f'