Khmer Optical Character Regconition
OCR technology has revolutionized identity verification, yet reliable software for Khmer identity documents is still lacking. Developing a Khmer OCR would enhance the KYC process and increase service access for Khmer speakers.
Our Motivation
Our AI team has achieved significant breakthroughs in developing various APIs that utilize advanced machine learning algorithms and natural language processing techniques to extract highly accurate digital text format from image of identity documents
Khmer OCR Pipeline
To develop an end-to-end OCR pipline for multi-font Khmer text recognition utilizing a deep learning-based sequence-to-sequrence model with attention mechanism.
Digitizing Khmer Document
By creating this Khmer OCR system, documents written in Khmer script can be easily digitized and processd. This can be especially helpful for business, organizations, and government agencies educational materials, and administrative paperwork.
Accuracy and Performance
To achieve the state-of-art performance in Khmer text recognition.
Our Features
OCR has always been our main driver. We built our model from scratch starting with National ID Card Detection. After months of developer, our team has tried our best crafted what seem to be impossible before. We offer some of the most advance OCR products on the market.
National ID Card
Extract information from an image of the National Identity Card of Cambodia in both English and Khmer languages
Internation Passport
Extract information from an image of the International Passport (Information in English only)
National Driving License
Extract information from an image of the Driving Licence of Cambodia in both English and Khmer languages
ID Card Detection
Detect the image of the card, remove the background, and isolate only the card
Demostration
Capability
Accurately extract information from NID in digital text format
Achieve high accuracy in both Khmer and English languages
Limitation
- API may fail to detect NID in some instances
- Rare cases may result in a few incorrect characters in specific NID fields
Instruction
- Avoid a white or highly contrasting background with the NID image
- Place the NID on a flat surface for the photograph
- Ensure each side of the NID card has a minimum of 300 pixels of background
ID Card Sample
