You need to enable JavaScript to run this app.
优惠活动
大模型
产品
解决方案
定价
更多

提升pytesseract识别精度时如何避免文本行丢失?

Tesseract/pytesseract OCR 行丢失问题处理建议

环境与代码

使用的版本信息:

tesseract 5.3.0.20221222
pytesseract.git@1ec07b1c9411ca58e5e3f05b0100f6bb0e8447ce

当前运行的Python代码:

import pytesseract
from pytesseract import pytesseract

pytesseract.tesseract_cmd = r'C:\\Program Files\\Tesseract-OCR\\tesseract.exe'
image = "C:\\ocr\\target\\31832_226140__0001-00004.jpg"

custom_config = r'--dpi 72 --oem 3 --psm 6 -c tessedit_char_whitelist="abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789.,\'\/- "'
pytestFolder = 'C:\\ocr\\increment_testing\\'

with open('{fn}debugTestLang1downAdjUp.txt'.format(fn=pytestFolder),'w') as f:
    f.write(pytesseract.image_to_string(image, lang='eng', config=custom_config))

图像情况

待处理的是72ppi高对比度灰度历史文档(1981年选民名册),图像参数:

  • 数据类型:uint8
  • 尺寸:2537×3640
  • 位深度:2

已完成的预处理

已实现以下预处理步骤,但未解决核心问题:

  • 拆分裁剪为独立页面和列
  • 校正图像歪斜(左侧页面存在轻微倾斜)
  • 调整图像尺寸
  • 高斯模糊、Otsu阈值处理

不同处理方式的OCR结果差异

  1. 原始图像直接识别:能识别所有文本行,但字符错误较多,结果示例:
REFERENCE ONLY
ALLAN-ANKER PENCARROW COMPOSITE ROLL PAGE 2 PAGE 3 PENCARROW COMPOSITE ROLL ANKER-ATKINSON
001 ALLAN, Wehom ROSS oo... 13 Kowhai Street, Wainuiomata,Greenkeeper 101 ANDERSON, Jean Elaine .........sseecreceeececeeteeeessss 1 Chanel Grove,Lower Clerk ANKER.
002 ALLARDYCE. Margaret isabel ..............89 Bell Roed,Lower Hutt,Housewife 102 ANDERSON, Joanatte Eo. crcsrncnvn. 16 Burden TuWencanas ig cerry pedi rans cacy ear ter iy Poor linn silted aga nen aR gle lated So st
003 ALLARDYCE, Mervyn George ............--..49 Bell Road, Lower Hutt,Company Mrgr 103 ANDERSON, Jeffrey Wilfred ....................288 Webwhetu Rd,Lower Hut.Bnk Clk FM 003 ANKER. Lewis Richard .......................4 Bledisloe Cr, Wainucomate Sales Clerk NAMA Sea ee J Donnelley Dave Wesnuiomata,Pub, Servent
004 ALLCOCK, George Mckinney ....s..n--.-0.41 Peel Place, Wainuiomata Scientist 104 ANDERSON, Jessie Faith ...................1/220 Warwhetu Road.Lower Hutt. Retired 004 ANKER, Robert James 368 We Rosa i HDR AOIY 'oorctoenincineed Donnelley Drive. Weinuiometa, Saleewomen
005 ALLCOCK. Shittey Constance 41 Peel Place, Weinuiomata Married 105 ANDERSON, John Gilbert... -1s.cnsrsenenn 65 Riverside Dr S,Lower Hutt,Tex Whr ae ARURRLEN se esse ee LINEAR VR gga 5 Wememiceter Roe W/rauerrara her ere
  1. 拆分单页并提升分辨率:字符识别精度提升,但会随机丢失文本行(丢失的并非质量较差的行),结果示例:
ALLAN-ANKER PENCARROW COMPOSITE ROLL PAGE 2
001 ALLAN, Witham ROSS ...u.......cs.ccsses 13 Kowhai Street, Wainuiomata,Greenkesper 101 ANDERSON, Jean Elaine .............0.csecssccecsseesse I Chanel Grove,Lower Hutt, Clerk
002 ALLARDYCE, Margaret Isabel ............................49 Bell Road,Lower Hutt, Housewife 102 ANDERSON, Jesnatte E ......eeccssn.s1 168 Burden Av,Weinuiomat gi Sorry ff
003 ALLARDYCE, Mervyn George .....................49 Bell Road,Lower Hutt,Company Mngr 103 ANDERSON, Jeffrey Wilfred ...........00.0-----288 Waiwhetu Rd,Lower Hutt.Bnk Clik
006 ALLEN, Albert Robert 0... 48 Coast Road, Wainuiomate Retired 106 ANDERSON, Joyce Lequesnoy ........................ 18 Bumside St,Lower Hutt. Housewife
007 ALLEN. Anthony Besumont ...........6.0-.--37 Kotan Rosd,Days Bry,Accounts Ofr 107 ANDERSON, Judith .................-............81 Waiwhetu Road,Lower Hutt,Housekeeper ff
008 ALLEN, Anthony James... ..e00.--27 Dover Road,WainuiomataElectro-Frmr 108 ANDERSON, Kathleen Mary ..........................27 Rossiter Avenue,Lower HuttRetired
O10 ALLEN, Colleen Mary .scccscsccseesecseereseniee-s fi Hine Road, Wainuiomata.Admin Asst. 110 ANDERSON, Linda Vicki ............-..s00.0.12 Kaun Street,Eastbourne,Mother Wife 5
  1. 裁剪为单列并提升分辨率:不同裁剪版本结果存在差异,比如单页识别中丢失的条目005在此种处理下能被识别。

当前需求

目前计划通过对双页、单页、单列等多种图像编辑版本分别进行OCR,再整合所有文本行来获取完整内容,但希望找到无需整合结果、提升分辨率同时不丢失文本行的解决方案。


内容的提问来源于stack exchange,提问作者Dave

相关产品推荐
方舟 Agent Plan

超全模态模型 × Harness 升级,最新支持 Deepseek-V4.1-Flash、GLM-5.3 系列、Doubao-Seedream-5.0-pro、Kimi-K3 (部分), 限时 9.9 元起

最近更新时间:2026.07.22 10:07:02