from __future__ import annotations import json import subprocess import xml.etree.ElementTree as ET from pathlib import Path from PIL import Image from translate_evans_ch2 import is_header_or_footer, line_kind ROOT = Path(__file__).resolve().parent PDF = ROOT / "Partial Differential Equations, Second Edition (Lawrence C. Evans) (z-library.sk, 1lib.sk, z-lib.sk).pdf" RAW = ROOT / "evans_chapter2_raw.txt" XML = ROOT / "evans_chapter2_bbox.xml" PAGE_DIR = ROOT / "evans_formula_pages_288" CROP_DIR = ROOT / "evans_formula_crops" MANIFEST = ROOT / "evans_formula_crops.json" GROUP_MANIFEST = ROOT / "evans_formula_group_crops.json" XML_GROUP_MANIFEST = ROOT / "evans_formula_xmlgroup_crops.json" PDF_FIRST_PAGE = 34 PDF_LAST_PAGE = 105 def run(cmd: list[str]) -> None: subprocess.run(cmd, cwd=ROOT, check=True) def ensure_xml() -> None: if XML.exists() and XML.stat().st_size > 1000: return with XML.open("wb") as f: subprocess.run( ["pdftohtml", "-xml", "-f", str(PDF_FIRST_PAGE), "-l", str(PDF_LAST_PAGE), "-i", "-stdout", str(PDF)], cwd=ROOT, check=True, stdout=f, ) def ensure_pages() -> None: PAGE_DIR.mkdir(exist_ok=True) marker = PAGE_DIR / f"page-{PDF_FIRST_PAGE}.png" if marker.exists(): return run([ "pdftoppm", "-png", "-r", "288", "-f", str(PDF_FIRST_PAGE), "-l", str(PDF_LAST_PAGE), str(PDF), str(PAGE_DIR / "page"), ]) def page_image(page_no: int) -> Path: # pdftoppm pads page numbers by the width needed for the requested range. candidates = sorted(PAGE_DIR.glob(f"page-{page_no}*.png")) if not candidates: candidates = sorted(PAGE_DIR.glob(f"page-*{page_no}.png")) if not candidates: raise FileNotFoundError(page_no) return candidates[0] def raw_formula_positions() -> list[dict[str, int]]: positions: list[dict[str, int]] = [] raw_pages = RAW.read_text(encoding="utf-8", errors="replace").replace("\r\n", "\n").split("\f") for page_idx, page in enumerate(raw_pages): if page_idx > PDF_LAST_PAGE - PDF_FIRST_PAGE: break line_index = 0 for raw_line in page.split("\n"): line = raw_line.rstrip() if is_header_or_footer(line) or not line.strip(): continue kind = line_kind(line) if kind == "formula": positions.append({"page": PDF_FIRST_PAGE + page_idx, "line_index": line_index}) line_index += 1 return positions def raw_formula_groups() -> list[dict[str, int]]: groups: list[dict[str, int]] = [] raw_pages = RAW.read_text(encoding="utf-8", errors="replace").replace("\r\n", "\n").split("\f") for page_idx, page in enumerate(raw_pages): if page_idx > PDF_LAST_PAGE - PDF_FIRST_PAGE: break page_no = PDF_FIRST_PAGE + page_idx line_index = 0 current: dict[str, int] | None = None for raw_line in page.split("\n"): line = raw_line.rstrip() if is_header_or_footer(line): continue if not line.strip(): continue kind = line_kind(line) if kind == "formula": if current is None: current = {"page": page_no, "start": line_index, "end": line_index} else: current["end"] = line_index else: if current is not None: groups.append(current) current = None line_index += 1 if current is not None: groups.append(current) return groups def xml_lines() -> dict[int, list[dict[str, float]]]: tree = ET.parse(XML) pages: dict[int, list[dict[str, float]]] = {} for page_el in tree.findall("page"): page_no = int(page_el.attrib["number"]) groups: dict[int, list[ET.Element]] = {} for text_el in page_el.findall("text"): txt = "".join(text_el.itertext()).strip() if not txt: continue top = int(float(text_el.attrib["top"])) key = round(top / 4) * 4 groups.setdefault(key, []).append(text_el) lines: list[dict[str, float]] = [] for _, elems in sorted(groups.items()): elems.sort(key=lambda e: float(e.attrib["left"])) txt = " ".join("".join(e.itertext()).strip() for e in elems).strip() if not txt or is_header_or_footer(txt): continue left = min(float(e.attrib["left"]) for e in elems) top = min(float(e.attrib["top"]) for e in elems) right = max(float(e.attrib["left"]) + float(e.attrib["width"]) for e in elems) bottom = max(float(e.attrib["top"]) + float(e.attrib["height"]) for e in elems) lines.append({"text": txt, "left": left, "right": right, "top": top, "bottom": bottom}) pages[page_no] = lines return pages def crop_formulas() -> None: ensure_xml() ensure_pages() CROP_DIR.mkdir(exist_ok=True) positions = raw_formula_positions() lines_by_page = xml_lines() manifest: list[str] = [] for i, pos in enumerate(positions): page_no = pos["page"] lines = lines_by_page.get(page_no, []) if pos["line_index"] >= len(lines): manifest.append("") continue line = lines[pos["line_index"]] img_path = page_image(page_no) with Image.open(img_path) as im: sx = im.width / 1188.0 sy = im.height / 1512.0 # Keep the full body width so equation numbers, side conditions, and limits survive. x0 = int(245 * sx) x1 = int(925 * sx) y0 = max(0, int((line["top"] - 10) * sy)) y1 = min(im.height, int((line["bottom"] + 12) * sy)) crop = im.crop((x0, y0, x1, y1)) out = CROP_DIR / f"formula_{i:04d}.png" crop.save(out) manifest.append(out.relative_to(ROOT).as_posix()) MANIFEST.write_text(json.dumps(manifest, ensure_ascii=False, indent=2), encoding="utf-8") group_manifest: list[str] = [] for i, group in enumerate(raw_formula_groups()): page_no = group["page"] lines = lines_by_page.get(page_no, []) if group["start"] >= len(lines) or group["end"] >= len(lines): group_manifest.append("") continue first = lines[group["start"]] last = lines[group["end"]] img_path = page_image(page_no) with Image.open(img_path) as im: sx = im.width / 1188.0 sy = im.height / 1512.0 x0 = int(245 * sx) x1 = int(925 * sx) y0 = max(0, int((first["top"] - 14) * sy)) y1 = min(im.height, int((last["bottom"] + 16) * sy)) crop = im.crop((x0, y0, x1, y1)) out = CROP_DIR / f"formula_group_{i:04d}.png" crop.save(out) group_manifest.append(out.relative_to(ROOT).as_posix()) GROUP_MANIFEST.write_text(json.dumps(group_manifest, ensure_ascii=False, indent=2), encoding="utf-8") xml_group_manifest: list[str] = [] xml_groups: list[dict[str, float]] = [] for page_no, lines in sorted(lines_by_page.items()): current: dict[str, float] | None = None for line in lines: txt = line.get("text", "") is_formula = line_kind(txt) == "formula" if is_formula: if current is None: current = { "page": page_no, "left": line["left"], "right": line["right"], "top": line["top"], "bottom": line["bottom"], } else: current["left"] = min(current["left"], line["left"]) current["right"] = max(current["right"], line["right"]) current["bottom"] = line["bottom"] else: if current is not None: xml_groups.append(current) current = None if current is not None: xml_groups.append(current) for i, group in enumerate(xml_groups): page_no = int(group["page"]) img_path = page_image(page_no) with Image.open(img_path) as im: sx = im.width / 1188.0 sy = im.height / 1512.0 x0 = int(245 * sx) x1 = int(925 * sx) y0 = max(0, int((group["top"] - 24) * sy)) y1 = min(im.height, int((group["bottom"] + 30) * sy)) crop = im.crop((x0, y0, x1, y1)) out = CROP_DIR / f"formula_xmlgroup_{i:04d}.png" crop.save(out) xml_group_manifest.append(out.relative_to(ROOT).as_posix()) XML_GROUP_MANIFEST.write_text(json.dumps(xml_group_manifest, ensure_ascii=False, indent=2), encoding="utf-8") print( f"wrote {len(manifest)} formula crops, " f"{len(group_manifest)} raw groups, and {len(xml_group_manifest)} xml groups" ) if __name__ == "__main__": crop_formulas()