Handwriting Datasets

The 15 Best Open-source Handwriting Datasets to Train your ML models

The business world is transforming at a phenomenal pace, yet this digital transformation is not nearly as wide-ranging as we would like it to be. People are still handling physical documents in their day-to-day operations, from large corporations to small-scale businesses. Although the frequency of usage has reduced considerably, it hasn’t completely been done away with. Instead of the time-consuming process of scanning documents for digital use, using the latest OCR is time-efficient and effective.

The rise in optical character recognition usage can primarily be attributed to the increase in the production of automatic recognition systems. As a result, the global market value of OCR technology, pegged at $8.93 billion in 2021, is predicted to grow at a CAGR of 15.4% between 2022 and 2030.

But what exactly is OCR technology? And why is it a game changer for businesses developing efficient AI models? Let’s find out.

What is OCR?

Alternatively referred to as text recognition, OCR or Optical Character Recognition is a program that extracts printed or written data from scanned documents, image-only PDFs, and handwritten notes into a machine-readable format. The software takes out each letter from the image and combines them into words and sentences, thus making it easy to access and edit the documents digitally.

What are open-source datasets?

There are several places where OCR technology has great potential to be leveraged. Some places include the airport, eBook publishing, advertisements, banks, and supply chain systems. However, for the applications to serve their purpose, they need to be trained on project-specific Optical Character Recognition datasets.

The efficiency of the application depends largely on the dataset’s quality and the training methodology involved. However, finding quality digital and handwriting datasets is difficult for the application. So, many companies use open-source or free-to-use datasets instead of proprietary ones.

Benefits and Challenges of Open-Source Datasets

Businesses need to pit the benefits and challenges against each other to understand whether they must opt for free-to-use data for their ML applications.


  • The data is easily available to access. Because of data availability, the cost of developing the application is reduced significantly.
  • The time and effort spent collecting data for the application are significantly reduced as the dataset is readily available.
  • There is an abundance of community forums or help groups that help learn, adapt and optimize the dataset.
  • One of the major advantages of the open-source dataset is it doesn’t lay any restrictions on customization.
  •   Open-Source data is accessible to a large section of the population, making analysis and innovation possible without monetary barriers.


  • The data specific to the project is difficult to acquire. Additionally, there is a possibility of missing information and incorrect use of the available data.
  • Acquiring proprietary data takes time, and effort and is costly
  • While it might be easier to acquire data, knowledge and analysis cost might outweigh the initial advantage.
  • Other developers also make use of the same data to develop applications.
  • These datasets are highly vulnerable to security breaches, privacy, and consent.

15 Best Handwriting & OCR Datasets for Machine Learning

Open-Source Ocr Datasets

Many open-source datasets are available for text recognition application development. Some of the best 15 are

  1. The ICDAR Dataset

    International Conference for Document Analysis and Recognition has a repository of 229 training and 233 testing images, along with annotations. It acts as a benchmark for text detection evaluation.

  2. IIIT 5K-Word Dataset

    Taken from Google image search, IIIT 5K-word is a collection of words from signboards, billboards, number plates, and posters. It contains 5K cropped word images making it one of the most extensive collections of text recognition datasets available.

  3. NIST Database

    The NIST or the National Institute of Science offers a free-to-use collection of over 3600 handwriting samples with more than 810,000 character images

  4. MNIST Database

    Derived from NSIT’s Special Database 1 and 3, the MNIST database is a compiled collection of 60,000 handwritten numbers for the training set and 10,000 examples for the test set. This open-source database helps train models to recognize patterns while spending less time on pre-processing.

  5. Text Detection

    An open-source database, the Text Detection dataset contains about 500 indoor and outdoor images of signboards, door plates, caution plates, and more.

  6. Stanford OCR

    Published by Stanford, this free-to-use dataset is a handwritten word collection by the MIT Spoken Language Systems Group.

  7. DDI-100

    Otherwise called the Distorted Document Images Dataset, the DDI-100 is a collection of over 6658 pages of documents with several geometric patterns and distortions applied. In addition, the DDI-100 has more than 99870 images, stamp masks, text masks, and bounding boxes.

  8. RoadText-1K

    One of the largest datasets that help train models to detect text in videos, the RoadText-1K contains 1000 video clips complete with bounding box text annotation and transcription of the text in every video frame.

  9. MSRA-TD500

    Contains 300 training and 200 text images; the MSRA-TD500 contains characters from Chinese and English languages and is annotated at the sentence level.

  10. MJSynth Dataset

    Provided by the University of Oxford, this word dataset has nearly 9 million synthetically generated images covering more than 90 thousand English language words.

  11. Street View Text

    Gathered from Google Street View images, this dataset has text detection images mainly of boards and street-level signs.

  12. Document Database

    The Document Database is a collection of 941 handwritten documents, including tables, formulas, drawings, diagrams, lists, and more, from 189 writers.

  13. Mathematics Expressions

    The Mathematics Expressions is a database that contains 101 mathematical symbols and 10,000 expressions.

  14. Street View House Numbers

    Harvested from Google Street View, this Street View House Numbers is a database containing 73257 street house number digits.

  15. Natural Environment OCR

    The Natural Environment OCR, is a dataset of nearly 660 images worldwide and 5238 text annotations.

These were some of the top open-source datasets for training ML models for text detection applications. Selecting the one that aligns with your business and application needs could take time and effort. However, you must experiment with these datasets before deciding on the appropriate one.

To help you progress toward a reliable and efficient text detection application is Shaip – the high-ranking technology solutions provider. We leverage our tech experience to create customizable, optimized, and efficient OCR training datasets for various client projects. To fully understand our capabilities, get in touch with us today.

Social Share