Project Description
Task: Handwritten OCR and Coordinate Extraction for Chinese Language Exam Papers
We need a developer to build an OCR module for Chinese language exam papers.
The module should take an image of a Chinese language exam paper and return the student’s handwritten Chinese answers with exact position coordinates.
Requirements:
1. Image cleanup
- Correct tilted or skewed photos
- Remove shadows and uneven lighting
- Reduce noise, stains, wrinkles, and background marks
- Improve handwriting clarity
2. Remove printed content
The system should ignore printed parts of the Chinese language exam paper, including:
- Question text
- Reading passages
- Grid lines
- Answer lines
- Borders
Only student handwriting should be recognized.
3. Handwritten Chinese recognition
- The system should support:
- Chinese handwriting
- Messy student handwriting
- Chinese essays
- Short-answer questions
- Open-ended Chinese language questions
4. Coordinate output
The system must return:
- Each line of text with its bounding box
- Each Chinese character with its own bounding box
Example output:
{
"lines": [
{
"text": "今天天气很好",
"bbox": [x1, y1, x2, y2],
"words": [
{"char": "今", "bbox": [x1, y1, x2, y2]},
{"char": "天", "bbox": [x1, y1, x2, y2]}
]
}
]
}
Important:
- Every recognized character must have a coordinate box
- Text and coordinates must match one by one
- Printed exam content should not be included
- Reading order must be correct
Input:
JPG, PNG, or scanned Chinese language exam paper image
Output:
JSON with handwritten text and coordinates
Testing:
We will test with:
- Chinese language exam papers
- Phone photos
- Tilted papers
- Uneven lighting
- Messy handwriting
- Scanned papers
Expected result:
- Around 90%+ handwriting recognition accuracy
- Coordinates should be accurate enough for marking on the original paper
Deliverables:
- Working API
- Source code
- Deployment instructions
- Simple test demo
- Must support private/local deployment