The solution addresses the following issues in the industry:
- Conventional OCR solutions target printed characters and are limited to only certain fonts and Latin languages.
- Recognition performance on handwritten characters in unlimited number of writing styles is subpar compared to printed characters.
- Inability to recognize Chinese language well with full coverage of its characters and their variants limits market potential.
- Usage of handwritten forms is still prominent in the financial and public sector.
- Recognition is usually done on the cloud, which requires persistent online connection, potentially jeopardizing the security of sensitive data.
Intelligent Character Recognition
The Intelligent Character Recognition Engine is a deep learning-based OCR engine and is one of the most essential components of our platform.
The engine is capable of classifying at least 8000 alphanumeric, Traditional and Simplified Chinese characters, which covers 漢語大字典, 小學常用字表, 小學教學參考詞語表 (試用) and most special Chinese characters that exist only in Hong Kong addresses, with high accuracy:
- Simplified Chinese accuracy: 97.0%*
- Traditional Chinese accuracy: 98.0%
- Numeric digits accuracy: 99.7%
- English alphabets accuracy: 99.0%
Handwritten Chinese Recognition
The Single-line Handwritten Chinese Recognition Engine leverages our Intelligent Character Recognition Engine and our character segmentation algorithm to perform OCR on a Chinese sentence, inheriting all features presented in our Intelligent Character Recognition Engine.
Handwritten English Recognition
Our Handwritten English Recognition Engine is a context-aware OCR engine that takes sentence context into account to further increase recognition accuracy. It supports both printed and handwritten alphanumeric text.
Our Digit-string OCR Engine offers robust recognition of pure numeric characters that exist in cheques and financial documents, such as dates, prices and mathematical expressions.
Hong Kong Address Database
* Accuracy on Simplified Chinese is based on the International test set from ICDAR 2013; Accuracy on digits is based on the test set from International dataset MNIST; Accuracy on Traditional Chinese is based on in-house test set.
Our HKID Recognition system enables low-cost, contactless and fast recognition of both generations of HKID. The system can be set up with a low-end computing device and a simple webcam. All processing is done on-device with a number of validation steps to provide additional robustness.
Document Comparison Platform
The Document Comparison Platform addresses the labor-intensive process of comparing business critical documents. The platform detects changes between different versions of the document, and visualize the differences in a printer-friendly format. It is also deployed as a web service for cross platform compatibility.
Public media reported December 2018. Our OCR technology has drawn attention from different industries, praising its speed, accuracy and practicality.