From fb3bd3de029b0ec327c0b9aac1f1ec0aa8ac2a21 Mon Sep 17 00:00:00 2001
From: euvre <93761161+euvre@users.noreply.github.com>
Date: Mon, 1 Jun 2026 04:22:11 -0700
Subject: [PATCH] fix(deepdoc): add English caption patterns to fix missing
 figure/table numbering (#15481)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

### What problem does this PR solve?
## Problem

When parsing PDFs containing English figure/table captions (e.g. "Fig.
20", "Figure 20", "Table 20"), the `is_caption` method in
`TableStructureRecognizer` failed to recognize them as captions. This
caused figure numbering gaps in the parsed output (e.g. Fig. 19 → Fig.
21, skipping Fig. 20).

## Root Cause

The `is_caption` regex only matched Chinese caption formats:

```python
patt = [r"[图表]+[ 0-9:：]{2,}"]
```

When the layout recognizer also failed to assign a `caption` layout type
to a given text block, English captions were entirely missed.

## Fix

Added three case-insensitive English caption patterns to `is_caption` in
`deepdoc/vision/table_structure_recognizer.py`:

- `(?i)Fig\.?\s*\d+` — matches `Fig. 20`, `Fig 20`, `FIG. 20`, etc.
- `(?i)Figure\s+\d+` — matches `Figure 20`, `FIGURE 20`, etc.
- `(?i)Table\s+\d+` — matches `Table 20`, `TABLE 20`, etc.

## Files Changed

- `deepdoc/vision/table_structure_recognizer.py` — extended `is_caption`
regex patterns


- [x] Bug Fix (non-breaking change which fixes an issue)

Signed-off-by: noob <yixiao121314@outlook.com>
---
 deepdoc/vision/table_structure_recognizer.py | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/deepdoc/vision/table_structure_recognizer.py b/deepdoc/vision/table_structure_recognizer.py
index e0892c2d72..997bc84b62 100644
--- a/deepdoc/vision/table_structure_recognizer.py
+++ b/deepdoc/vision/table_structure_recognizer.py
@@ -112,7 +112,12 @@ class TableStructureRecognizer(Recognizer):
 
     @staticmethod
     def is_caption(bx):
-        patt = [r"[图表]+[ 0-9:：]{2,}"]
+        patt = [
+            r"[图表]+[ 0-9:：]{2,}",
+            r"(?i)Fig\.?\s*\d+",
+            r"(?i)Figure\s+\d+",
+            r"(?i)Table\s+\d+",
+        ]
         if any([re.match(p, bx["text"].strip()) for p in patt]) or bx.get("layout_type", "").find("caption") >= 0:
             return True
         return False