Enhanced Inhouse OCR Engine
OVERVIEW
The requirement was to design a system using only open source
REQUIREMENTS
- Build NATIVE(for OCR) APP iOS7+ platform
- APP will support only «English» language and English alphanumeric characters on a data plate
- Printed, engrained, embossed data plates
- After taking picture, with drag method, a rectangle to be drawn on the area of interest.
- Adapt and extend open source OCR library to meet the recognition and conversion of data plate –serial number(picture to text)
- Solution will work for reading «Data plate» for competition equipment, so it is future proof
- While picture is taken, APP will record its GPS location or GSM location
CHALLENGES
- Engraved/embossed images/metallic tags are low in contrast which becomes difficult for OCR to detect.
- Images taken with flash on or very bright surface shine gives false result.
- If too much extra area is selected, due to noise OCR recognizes false characters.
- Characters connected with each other is difficult to process for OCR.
- As locational logic (like vehicle license plates) is not applicable in alpha numeric strings of text, the similar appearance between numbers and characters like I & 1, 5 & S, O & 0, B & 8 etc. may lead to wrong recognition and would not be considered for success rate percentage
SOLUTIONS
- Developing an image rejection module that checks image quality & provides feedback to recapture image.
- Automatic skew detection and correction if text having some rotation.
- Flashy effect removal using homomorphic filter.
- Line removal, if text have some small linkage with line.
- Automatic text categorization into white and black text using image processing techniques.
- Separate image pre-processing modules considering black and white text.
- Enhanced OCR engine for specific metallic images.
