Deadline: May 04, 2018 by 24:00
Further Information: Your code is part of your delivery, so please make sure that your code is well-documented and as readable as possible. The first section of the report must contain an explanation on how to run your code.
- Source Code
- Report of approx. 4 pages
- Any additional material (e.g. plots, ...)
- Introduction: How to run the code
- Feature Engineering:
- Explain how each of the feature engineering techniques you selected work. Why did you select these techniques? Justify your answer.
- Were there any techniques that you wanted to try, but for some reason were not able to? Explain.
- Character Classification: Use at least 2 different models (NN is allowed!)
- After having looked at the dataset, did you have any initial idea about what kind of model that might work well? Explain your reasoning.
- Give a description of the two models you elected to use10. The description must include a brief explanation on how they work. Why did you select these models?
- A critical evaluation of your two models. How are you measuring their performance? How did they do? Which model gave the best results? Include at least five predictions in the report (both good and bad).
- Were there any additional models that you would have liked to try, but for some reason were not able to? Explain.
- Character Detection:
- Test your character detector on detection-1.jpg and detection-2.jpg and show the result in the report. Feel free to find or create additional images to test your detector, if you are so inclined.
- Give an evaluation of your detection system. How does it perform?
- Describe any improvements you made to your detector. Discuss how you can improve your system further.
- Conclusion:
- What is the weakest and strongest component of your OCR system (feature engineering, character classification, and character detection)? Explain your answer.
- What went good/bad with the project? Any lessons learned?