Update app.py

app.py

@@ -83,7 +83,6 @@ MIN_CROP_PIXELS = 112 * 112
 
 _parser = None
 
-
 def get_parser():
     global _parser
     if _parser is None:
@@ -91,7 +90,6 @@ def get_parser():
         _parser = GlmOcr(api_key=GLMOCR_API_KEY, mode="maas")
     return _parser
 
-
 # ---------------------------------------------------------------------------
 # Best-effort config tweaks (safe to fail on read-only HF env)
 # ---------------------------------------------------------------------------
@@ -123,7 +121,6 @@ try:
 except Exception:
     pass
 
-
 # --------------------------
 # Header/footer helpers
 # --------------------------
@@ -140,7 +137,6 @@ def get_header_footer_zones(regions, norm_height=1000):
         return None, None
     return min(y_tops) / norm_height, max(y_bottoms) / norm_height
 
-
 def extract_zone_text_pdf(pdf_path, page_num, y_start_frac, y_end_frac):
     try:
         import pymupdf as fitz
@@ -154,7 +150,6 @@ def extract_zone_text_pdf(pdf_path, page_num, y_start_frac, y_end_frac):
     except Exception:
         return ""
 
-
 def extract_pdf_text_in_band(pdf_path, page_num, y_start_frac, y_end_frac):
     try:
         import pymupdf as fitz
@@ -175,7 +170,6 @@ def extract_pdf_text_in_band(pdf_path, page_num, y_start_frac, y_end_frac):
     except Exception:
         return ""
 
-
 def ocr_zone(image_path, y_start_frac, y_end_frac):
     zone_name = "header" if y_end_frac < 0.5 else "footer"
     try:
@@ -221,7 +215,6 @@ def ocr_zone(image_path, y_start_frac, y_end_frac):
         log.warning("[%s] ocr_zone failed: %s", zone_name, e, exc_info=True)
     return ""
 
-
 def fix_account_number(hdr: str) -> str:
     if not hdr:
         return hdr
@@ -236,7 +229,6 @@ def fix_account_number(hdr: str) -> str:
             hdr = hdr[len(acct):].lstrip()
     return hdr
 
-
 # --------------------------
 # Table stabilization helpers
 # --------------------------
@@ -256,7 +248,6 @@ def close_unclosed_html(md: str) -> str:
             md += ("</%s>" % tag) * (opened - closed)
     return md
 
-
 def looks_like_markdown_table(block: str) -> bool:
     lines = [ln.rstrip() for ln in block.strip().splitlines() if ln.strip()]
     if len(lines) < 2:
@@ -266,7 +257,6 @@ def looks_like_markdown_table(block: str) -> bool:
     sep = lines[1].replace(" ", "")
     return ("---" in sep) and ("|" in sep)
 
-
 def md_table_to_html(block: str) -> str:
     lines = [ln.strip() for ln in block.strip().splitlines() if ln.strip()]
     if len(lines) < 2:
@@ -292,7 +282,6 @@ def md_table_to_html(block: str) -> str:
         html_rows.append("<tr>" + "".join(f"<td>{html.escape(c)}</td>" for c in cols[: len(header)]) + "</tr>")
     return "<table>\n" + "\n".join(html_rows) + "\n</table>"
 
-
 def normalize_money_glyphs(text: str) -> str:
     if not text:
         return text
@@ -310,7 +299,6 @@ def normalize_money_glyphs(text: str) -> str:
     t = re.sub(r"\b[0-9Oo\$,.\-]{4,}\b", o_to_zero, t)
     return t
 
-
 # ---- Generic HTML table normalizer ----
 class TableGridParser(HTMLParser):
     """Parse a <table> into rows of (text, colspan, rowspan)."""
@@ -346,7 +334,6 @@ class TableGridParser(HTMLParser):
         if self._in_cell:
             self._cell_text.append(data)
 
-
 def _build_grid(rows_data):
     if not rows_data:
         return []
@@ -372,7 +359,6 @@ def _build_grid(rows_data):
             row.append("")
     return grid
 
-
 def _grid_to_html(grid):
     if not grid:
         return ""
@@ -393,13 +379,11 @@ def _grid_to_html(grid):
     lines.append("</table>")
     return "\n".join(lines)
 
-
 def _is_amount_like(s: str) -> bool:
     if not s:
         return False
     return re.fullmatch(r"\$?-?\d{1,3}(?:,\d{3})*(?:\.\d{2})?", s.strip()) is not None
 
-
 def _drop_truly_empty_columns(grid):
     """
     Drop columns that are empty in header AND almost always empty in body.
@@ -438,7 +422,6 @@ def _drop_truly_empty_columns(grid):
         new_grid.append([cell for idx, cell in enumerate(row) if idx < len(keep) and keep[idx]])
     return new_grid
 
-
 def _merge_blank_header_text_columns(grid):
     """
     If the header row has blank columns, and most body rows have non-empty *text* in that
@@ -502,7 +485,6 @@ def _merge_blank_header_text_columns(grid):
         new_grid.append([cell for idx, cell in enumerate(row) if idx < len(keep) and keep[idx]])
     return new_grid
 
-
 # --------------------------
 # Shared keyword definitions
 # --------------------------
@@ -588,7 +570,6 @@ _HEADER_KW_EXACT_RE = re.compile(
 # to be considered a misplaced header row.
 _HEADER_ROW_KEYWORD_THRESHOLD = 0.5
 
-
 # --------------------------
 # Fix 1: Fused-header recovery
 # --------------------------
@@ -607,7 +588,6 @@ def _split_keyword_remainder(cell_text: str):
             return keyword, remainder
     return s, ""
 
-
 def _extract_fused_header_artifacts(header_row):
     """
     Generic detector for the OCR artifact where the first data row of a table
@@ -661,7 +641,6 @@ def _extract_fused_header_artifacts(header_row):
 
     return list(header_row), None
 
-
 def _clean_header_artifacts(grid):
     """
     Entry point for Fix 1 called from normalize_html_tables.
@@ -674,7 +653,6 @@ def _clean_header_artifacts(grid):
     grid[0] = cleaned_header
     return grid, recovered_row
 
-
 # --------------------------
 # Fix 2: Misplaced header row promotion
 # --------------------------
@@ -702,7 +680,6 @@ def _row_keyword_score(row):
 
     return keyword_hits / len(non_empty)
 
-
 # Pattern: header cell contains balance/account info — do NOT promote away from it.
 # Matches things like "Beginning Balance:", "Ending Balance:", account numbers, "$20.48$3.46"
 _BALANCE_INFO_RE = re.compile(
@@ -713,7 +690,6 @@ _BALANCE_INFO_RE = re.compile(
     re.IGNORECASE,
 )
 
-
 def _header_contains_balance_info(header_row):
     """
     Return True if any cell in the header row contains balance or account
@@ -725,7 +701,6 @@ def _header_contains_balance_info(header_row):
             return True
     return False
 
-
 def _promote_misplaced_header_row(grid):
     """
     Detects and fixes the OCR artifact where real column headers land in a
@@ -826,7 +801,6 @@ def _promote_misplaced_header_row(grid):
 
     return new_grid
 
-
 # --------------------------
 # Fix 3: Fused key-value rows in summary sections
 # --------------------------
@@ -840,7 +814,6 @@ _FUSED_KV_RE = re.compile(
     re.DOTALL,
 )
 
-
 def _fix_fused_keyvalue_rows(grid):
     """
     Fix rows where label+value are fused into the first cell with all other
@@ -930,7 +903,6 @@ def _fix_fused_keyvalue_rows(grid):
 
     return new_grid
 
-
 # --------------------------
 # Fix 4: PDF text-layer row-count patch
 # --------------------------
@@ -1074,6 +1046,40 @@ def _extract_textlayer_rows(pdf_path: str, page_num: int):
         return []
 
 
+def _jb_extract_checks_fallback(text_layer: str) -> List[Tuple[str, str, str]]:
+    """
+    Johnson Bank: when pdfminer column extraction fails (e.g. pymupdf line text or
+    multi-column layout), extract (date, check#, amount) triplets from the Checks
+    subsection. Anchored by 'Checks (' ... 'Daily Account Balance' only.
+    """
+    if not text_layer:
+        return []
+    m = re.search(
+        r"Checks\s*\([^\n]*\n(.*?)(?=^Daily\s+Account\s+Balance\s*$)",
+        text_layer,
+        re.DOTALL | re.MULTILINE | re.IGNORECASE,
+    )
+    if not m:
+        return []
+    chunk = m.group(1)
+    chunk = re.sub(
+        r"^\s*Date\s+Number\s+Amount\s*$",
+        "",
+        chunk,
+        flags=re.MULTILINE | re.IGNORECASE,
+    )
+    _triplet = re.compile(
+        r"(\d{2}-\d{2})\s+(\*?\d+)\s+(\d{1,3}(?:,\d{3})*\.\d{2})"
+    )
+    out: List[Tuple[str, str, str]] = []
+    seen = set()
+    for t in _triplet.findall(chunk):
+        if t not in seen:
+            seen.add(t)
+            out.append(t)
+    return out
+
+
 def _extract_jb_summary_sections(text_layer: str, needs_checks: bool = True, needs_dab: bool = True) -> str:
     """
     Parse Johnson Bank Checks and Daily Account Balance from pdfminer column-separated
@@ -1141,6 +1147,9 @@ def _extract_jb_summary_sections(text_layer: str, needs_checks: bool = True, nee
                 break
             i += 1
 
+        if not checks and re.search(r"\bChecks\s*\(", text_layer, re.I):
+            checks = _jb_extract_checks_fallback(text_layer)
+
         if checks:
             rows = "\n".join(
                 "<tr><td>" + d + "</td><td>" + n + "</td><td>" + a + "</td></tr>"
@@ -1148,7 +1157,7 @@ def _extract_jb_summary_sections(text_layer: str, needs_checks: bool = True, nee
             )
             output.append("Checks")
             output.append(
-                "<table>\n<tr><th>Date</th><th>Number</th><th>Amount</th></tr>\n"
+                '<table class="jb-summary-checks">\n<tr><th>Date</th><th>Number</th><th>Amount</th></tr>\n'
                 + rows + "\n</table>"
             )
 
@@ -1429,8 +1438,6 @@ def _patch_ocr_with_textlayer(page_md: str, pdf_path: str, page_num: int) -> str
         log.warning("_patch_ocr_with_textlayer failed (page %d): %s", page_num, e)
         return page_md  # always safe — return original on any error
 
-
-
 # --------------------------
 # Fix 5: Mid-table separator row reconstruction
 # --------------------------
@@ -1443,7 +1450,6 @@ _SPLIT_AMOUNT_RE = re.compile(
     r"^-?\$?\d{1,3}(?:,\d{3})*(?:\.\d{1,4})?$|^\$-?\d{1,3}(?:,\d{3})*(?:\.\d{1,4})?$"
 )
 
-
 def _reconstruct_separator_rows(grid):
     """
     Detect and reconstruct mid-table separator / label rows that OCR has
@@ -1643,7 +1649,6 @@ def _merge_split_rows(grid):
 
     return new_grid
 
-
 def _extract_fused_desc_amount(grid):
     """
     Fix OCR artifact where a trailing money amount gets fused into the
@@ -1707,7 +1712,6 @@ def _extract_fused_desc_amount(grid):
 
     return new_grid
 
-
 def _is_junk_table(grid):
     """
     Detect and suppress known junk tables that are OCR artifacts from
@@ -1732,7 +1736,6 @@ def _is_junk_table(grid):
             return True
     return False
 
-
 def _split_fused_multicolumn_header(grid):
     """
     Fix OCR artifact where multiple column headers are fused into one <th> cell.
@@ -1851,7 +1854,6 @@ def _split_fused_multicolumn_header(grid):
 
     return new_grid
 
-
 def _normalize_daily_balance_table(grid):
     """
     Fix OCR artifact in DAILY BALANCE SUMMARY tables where:
@@ -1929,7 +1931,6 @@ def _normalize_daily_balance_table(grid):
 
     return [new_header] + new_rows
 
-
 def _normalize_checks_paid_table(grid):
     """
     Fix OCR artifact in CHECKS PAID SUMMARY tables where:
@@ -2007,7 +2008,6 @@ def _normalize_checks_paid_table(grid):
 
     return new_rows
 
-
 # ── Johnson Bank plain-text section converter ─────────────────────────────
 # Johnson Bank formats Deposits, Withdrawals, Checks, and Daily Account Balance
 # as plain text columns (no HTML tables). This converter detects those sections
@@ -2035,7 +2035,6 @@ _JB_DATE_DD  = re.compile(r"^\d{2}-\d{2}\s+")
 _JB_TXN_ROW  = re.compile(r"^(\d{2}-\d{2})\s+(.+?)\s+(-?\d{1,3}(?:,\d{3})*\.\d{2})\s*$")
 _JB_BAL_ROW  = re.compile(r"^(\d{2}-\d{2})\s+(\d{1,3}(?:,\d{3})*\.\d{2})\s*$")
 
-
 def _jb_rows_to_html(header_cols, rows):
     th    = "".join("<th>" + h + "</th>" for h in header_cols)
     parts = ["<table>", "<tr>" + th + "</tr>"]
@@ -2045,12 +2044,10 @@ def _jb_rows_to_html(header_cols, rows):
     parts.append("</table>")
     return "\n".join(parts)
 
-
 def _jb_section_has_table(lines, start, end):
     """Return True if any line in lines[start:end] contains a <table tag."""
     return any("<table" in l.lower() for l in lines[start:end])
 
-
 def _find_first_data_row(lines, start):
     """Return index of first non-blank line after start, or None."""
     for i in range(start, min(start + 20, len(lines))):
@@ -2058,7 +2055,6 @@ def _find_first_data_row(lines, start):
             return i
     return None
 
-
 def convert_plaintext_bank_sections(text: str) -> str:
     """
     Convert Johnson Bank plain-text transaction sections to HTML tables.
@@ -2192,7 +2188,6 @@ def convert_plaintext_bank_sections(text: str) -> str:
 
     return "\n".join(out)
 
-
 def normalize_html_tables(text: str) -> str:
     """
     For every <table>...</table>:
@@ -2269,7 +2264,6 @@ def normalize_html_tables(text: str) -> str:
     out.append(text[last:])
     return "".join(out)
 
-
 def stabilize_tables_and_text(page_md: str) -> str:
     if not page_md:
         return page_md
@@ -2289,7 +2283,6 @@ def stabilize_tables_and_text(page_md: str) -> str:
     stabilized = normalize_html_tables(stabilized)
     return close_unclosed_html(stabilized)
 
-
 # --------------------------
 # PDF rendering with padding
 # --------------------------
@@ -2329,7 +2322,6 @@ def render_pdf_pages_to_images(pdf_path: str) -> Tuple[List[str], List[int]]:
     doc.close()
     return page_images, page_heights
 
-
 # --------------------------
 # GLM-OCR result extraction
 # --------------------------
@@ -2350,7 +2342,6 @@ def get_page_md_and_regions(page_result):
                 regions = r.get("regions") or []
     return md, regions
 
-
 # --------------------------
 # Main entry
 # --------------------------
@@ -2414,7 +2405,7 @@ def run_ocr(uploaded_file):
                 # Run OUTSIDE _patch_ocr_with_textlayer so exceptions there don't block it.
                 if is_pdf:
                     try:
-                        needs_checks = "checks" not in stabilized.lower() or "<th>Number</th>" not in stabilized
+                        needs_checks = 'class="jb-summary-checks"' not in stabilized
                         needs_dab    = "daily account balance" not in stabilized.lower()
                         if needs_checks or needs_dab:
                             # Try pdfminer first (column-aware), fall back to pymupdf
@@ -2495,7 +2486,6 @@ def run_ocr(uploaded_file):
             except Exception:
                 pass
 
-
 with gr.Blocks(title="GLM-OCR") as demo:
     gr.Markdown("# GLM-OCR\nUpload a PDF or image. Headers included; tables stabilized.")
     file_in = gr.File(label="Upload PDF or image", file_types=[".pdf", ".png", ".jpg", ".jpeg", ".tiff", ".bmp"])