From 9d397a9db40774439a6fcdcf992d3f109ceccc74 Mon Sep 17 00:00:00 2001
From: Chia-Hsuan Hsu <winnie920924@gmail.com>
Date: Mon, 4 May 2026 20:31:40 +0800
Subject: [PATCH] filter figure captions and URLs from heading detection

Add filtering in is_heading_candidate() to prevent
false positives:
- Figure/Table captions (e.g. "Figure 1: Results")
- URLs (http/https/www)
- Email addresses
---
 crates/open-core/src/pdf.rs | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/crates/open-core/src/pdf.rs b/crates/open-core/src/pdf.rs
index a8a98f7..81d0b06 100644
--- a/crates/open-core/src/pdf.rs
+++ b/crates/open-core/src/pdf.rs
@@ -644,6 +644,23 @@ fn is_heading_candidate(text: &str, is_first: bool) -> bool {
         return false;
     }
 
+    let lower = trimmed.to_lowercase();
+
+    if lower.starts_with("figure ") || lower.starts_with("fig.") || lower.starts_with("table ") {
+        return false;
+    }
+
+    if trimmed.starts_with("http://")
+        || trimmed.starts_with("https://")
+        || trimmed.starts_with("www.")
+    {
+        return false;
+    }
+
+    if trimmed.contains('@') && !trimmed.contains(' ') {
+        return false;
+    }    
+
     let len = trimmed.len();
 
     if is_first && len >= 3 {