From 7d5888c2f189fc77f6f768a1fe505b30fe02af5d Mon Sep 17 00:00:00 2001
From: H <43509927+guoyuhao2330@users.noreply.github.com>
Date: Thu, 11 Jul 2024 11:01:55 +0800
Subject: [PATCH 1/2] Update pdf_parser.py

---
 deepdoc/parser/pdf_parser.py | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/deepdoc/parser/pdf_parser.py b/deepdoc/parser/pdf_parser.py
index 5bb9bffac06..a12fc343da4 100644
--- a/deepdoc/parser/pdf_parser.py
+++ b/deepdoc/parser/pdf_parser.py
@@ -286,6 +286,20 @@ def __ocr(self, pagenum, img, chars, ZM=3):
             self.mean_height[-1] / 3
         )
 
+        # solve char content confusion
+        record_error_length = 0
+        for c in Recognizer.sort_X_firstly(chars, self.mean_width[pagenum - 1] // 4)[0:128]:
+            ii = Recognizer.find_overlapped(c, bxs)
+            if ii is None:
+                self.lefted_chars.append(c)
+                continue
+            record_error_length += abs((bxs[ii]["bottom"] + bxs[ii]["bottom"] - c["bottom"] - c["top"]) / 2)
+
+        record_error_length = record_error_length / 128
+        for char in chars:
+            char["top"] -= record_error_length
+            char["bottom"] -= record_error_length
+        
         # merge chars in the same rect
         for c in Recognizer.sort_X_firstly(
                 chars, self.mean_width[pagenum - 1] // 4):

From c5518a5329d59b25c20bb5e1df1d691fc3b7d0a8 Mon Sep 17 00:00:00 2001
From: H <43509927+guoyuhao2330@users.noreply.github.com>
Date: Thu, 11 Jul 2024 12:07:00 +0800
Subject: [PATCH 2/2] Update pdf_parser.py

---
 deepdoc/parser/pdf_parser.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/deepdoc/parser/pdf_parser.py b/deepdoc/parser/pdf_parser.py
index a12fc343da4..fce9e91de4c 100644
--- a/deepdoc/parser/pdf_parser.py
+++ b/deepdoc/parser/pdf_parser.py
@@ -288,10 +288,9 @@ def __ocr(self, pagenum, img, chars, ZM=3):
 
         # solve char content confusion
         record_error_length = 0
-        for c in Recognizer.sort_X_firstly(chars, self.mean_width[pagenum - 1] // 4)[0:128]:
+        for c in chars[0:128]:
             ii = Recognizer.find_overlapped(c, bxs)
             if ii is None:
-                self.lefted_chars.append(c)
                 continue
             record_error_length += abs((bxs[ii]["bottom"] + bxs[ii]["bottom"] - c["bottom"] - c["top"]) / 2)