Neura Market
Back to .md Directory
BENCHMARKS.md

Benchmarks

Compares Tesseract OCR versions 3.05, 4.1, and 5.0 across three tessdata sets, showing execution times in seconds.

tesseract-ocr
May 2, 2026
0 downloads
1 views
View source

What this file does

Compares Tesseract OCR versions 3.05, 4.1, and 5.0 across three tessdata sets, showing execution times in seconds.

When to use it

  • Choosing between tessdata_best, tessdata_fast, or tessdata for your project
  • Deciding whether to build Tesseract with AVX/SSE4 or OpenMP support
  • Evaluating performance differences between Tesseract 3.x, 4.x, and 5.x

Assumes this stack

TesseractPythonpytesseracttimeit

Benchmarks

This page is dedicated to simple benchmarking of various tesseract version and options. As input image for testing is used image from issue 236.

Results

Buildtessdata_best tessdata_fasttessdata
305--2.4713
413noavx37.60525.158910.1519
413avx12.73002.95384.0860
5016.19812.12412.9107
501ap6.13692.12542.9221
501openmp3.45901.96122.3554

Information about testing environment

  • Windows 10 64bit
  • compiler: VS 2019
  • CPU: Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz 6 cores
  • memory: 16GB RAM

Tested with python code

import timeit
import time
import os
import pytesseract

start_time = time.time()
tess_exe = r"msvc.v5.openmp\tesseract.exe"
test_image = r"i263_speed.jpg"
os.environ['TESSDATA_PREFIX'] = r"tessdata_best\tessdata"

code_to_test = """
import pytesseract
pytesseract.pytesseract.tesseract_cmd = r"{}"
pytesseract.pytesseract.image_to_string(r"{}", lang = 'eng')
"""

runs = 15
elapsed_time = timeit.timeit(code_to_test.format(tess_exe, test_image), number=runs)/runs
print("\nDuration:", elapsed_time)

Tesseract build info

information provided by tesseract -v

3.05

305

It uses the legacy engine.

tesseract 3.05.02
 leptonica-1.83.0 (Dec 17 2021, 17:33:37) [MSC v.1929 LIB Release x64]
  libgif 5.2.1 : libjpeg 6b (libjpeg-turbo 2.0.91) : libpng 1.6.37 : libtiff 4.3.0 : zlib 1.2.11 : libwebp 1.2.0 : libopenjp2 2.4.0

4.1

413noavx

Build without AVX2/AVX/SSE4 support

tesseract 4.1.3
 leptonica-1.83.0 (Dec 17 2021, 17:33:37) [MSC v.1929 LIB Release x64]
  libgif 5.2.1 : libjpeg 6b (libjpeg-turbo 2.0.91) : libpng 1.6.37 : libtiff 4.3.0 : zlib 1.2.11 : libwebp 1.2.0 : libopenjp2 2.4.0
 Found libarchive 3.5.1 zlib/1.2.11 liblzma/5.2.4 bz2lib/1.0.6 libzstd/1.4.9

413avx

Build with AVX2/AVX/SSE4 support

tesseract 4.1.3-1-ge9986
 leptonica-1.83.0 (Jan 26 2022, 19:15:03) [MSC v.1929 LIB Release x64]
  libgif 5.2.1 : libjpeg 6b (libjpeg-turbo 2.0.91) : libpng 1.6.37 : libtiff 4.3.0 : zlib 1.2.11
 Found AVX2
 Found AVX
 Found FMA
 Found SSE
 Found libarchive 3.5.1 zlib/1.2.11 liblzma/5.2.4 bz2lib/1.0.6 libzstd/1.4.9

5.0

501

AVX2 supported

tesseract 5.0.1
 leptonica-1.83.0 (Dec 17 2021, 17:33:37) [MSC v.1929 LIB Release x64]
  libgif 5.2.1 : libjpeg 6b (libjpeg-turbo 2.0.91) : libpng 1.6.37 : libtiff 4.3.0 : zlib 1.2.11 : libwebp 1.2.0 : libopenjp2 2.4.0
 Found AVX2
 Found AVX
 Found FMA
 Found SSE4.1
 Found libarchive 3.5.1 zlib/1.2.11 liblzma/5.2.4 bz2lib/1.0.6 libzstd/1.4.9
 Found libcurl/7.75.0 zlib/1.2.11 libssh2/1.10.1_DEV

501ap

build with: cmake -E env CXXFLAGS="/Qpar /fp:fast" cmake ..

tesseract 5.0.1
 leptonica-1.83.0 (Dec 17 2021, 17:33:37) [MSC v.1929 LIB Release x64]
  libgif 5.2.1 : libjpeg 6b (libjpeg-turbo 2.0.91) : libpng 1.6.37 : libtiff 4.3.0 : zlib 1.2.11 : libwebp 1.2.0 : libopenjp2 2.4.0
 Found AVX2
 Found AVX
 Found FMA
 Found SSE4.1
 Found libarchive 3.5.1 zlib/1.2.11 liblzma/5.2.4 bz2lib/1.0.6 libzstd/1.4.9
 Found libcurl/7.75.0 zlib/1.2.11 libssh2/1.10.1_DEV

501openmp

The OpenMP build is known to waste a lot of CPU time. As several users report problems, it is disabled by default in version 5.0.1 and later releases. For other versions (>= 4.x) it is suggested to use the environment variable OMP_THREAD_LIMIT=1. Input from OpenMP experts would be appreciated.

tesseract 5.0.1
 leptonica-1.83.0 (Dec 17 2021, 17:33:37) [MSC v.1929 LIB Release x64]
  libgif 5.2.1 : libjpeg 6b (libjpeg-turbo 2.0.91) : libpng 1.6.37 : libtiff 4.3.0 : zlib 1.2.11 : libwebp 1.2.0 : libopenjp2 2.4.0
 Found AVX2
 Found AVX
 Found FMA
 Found SSE4.1
 Found OpenMP 2019
 Found libarchive 3.5.1 zlib/1.2.11 liblzma/5.2.4 bz2lib/1.0.6 libzstd/1.4.9
 Found libcurl/7.75.0 zlib/1.2.11 libssh2/1.10.1_DEV

What's inside

1 results table, 1 testing environment section, 1 Python benchmarking script, 6 build info blocks

Change this for your project

  • Replace msvc.v5.openmp\tesseract.exe with your own Tesseract executable path
  • Replace i263_speed.jpg with your own test image filename
  • Replace tessdata_best\tessdata with your own tessdata directory path

Where it goes

Keep it in your repository where the agent or team that needs it will read it.

Worth borrowing

  • Benchmarking multiple build variants side by side in a single table
  • Using Python's timeit for consistent timing across runs

Related Documents

BENCHMARKS.md

Ruby 2.7

Documents every language, core class, and standard library change in Ruby 2.7 with code examples and rationale.

ai
0
7
rubyreferences
BENCHMARKS.md

OABench: Benchmarking Large Language Models on the Brazilian Bar Examination

Evaluates 11 LLMs on the Brazilian Bar Exam's first phase, reporting accuracy, cost, and latency across three exam editions.

aillmeval
0
7
robertotcestari
BENCHMARKS.md

Prometheus Automation AI Marketplace - Project Documentation

Documents an enterprise AI marketplace built with Next.js 15, covering architecture, AI algorithms, security, and deployment.

aiworkflowautomation
0
4
Prometheus-Automation
BENCHMARKS.md

Benchmarks

Compares Okra's read/write performance against raw LMDB across three dataset sizes using small key-value pairs.

0
4
canvasxyz