https://github.com/and2797/optical_character_recognition-

Optical Character Recognition
https://github.com/and2797/optical_character_recognition-

Last synced: 2 months ago
JSON representation

Optical Character Recognition

• Host: GitHub
• URL: https://github.com/and2797/optical_character_recognition-
• Owner: AND2797
• Created: 2019-12-06T06:17:50.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2020-08-02T17:40:58.000Z (almost 6 years ago)
• Last Synced: 2025-01-06T03:41:52.289Z (over 1 year ago)
• Language: Python
• Size: 306 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
## Optical Character Recognition

Extract text from images and classify accordingly. 

### Assumptions

* Letters are reasonably spaced. 

* Letters should reasonably resemble the training data set. 

* Image should be naturally decipherable i.e free from any occlusions and obfuscations.

### Output 

![Output](https://raw.githubusercontent.com/AND2797/Optical_Character_Recognition-/master/Output.png)

### Observations

* Common Confusions -

  * True Label 'C' -> Predicted Label 'G'

  * True Label '1' -> Predicted Label - '|' (I)

  * True Label '4' -> Predicted Label - 'A'

  * True Label 'O'/ '0' -> Predicted Label - 'O'/'0'/'Q'

  * True Label '8' -> Predicted Label 'B'

  * True Label '6' -> Predicted Label 'G'

  * True Label '9' -> Predicted Label 'g' (single storey) 

  

  * Most of the confusions share highly similar features and construction, for example 4 and A, 8 and B, even C and G to some extent. 

### Tasks

- [X] Train (EMNIST - Balanced)

- [X] Extract text from images

- [X] Pre-process to match EMNIST 

- [X] Test on extracted

- [X] Improve accuracy and pre-processing

- [ ] Realtime OCR from Video