Denoising Dirty Documents
Objective
Clean the images of text (remove wrinkles, coffee stains, etc) to make them more readable by OCR.
Data
The data for this competition can be obtained through Kaggle.
Analysis
No Clean Benchmark
| Measure | Value |
|---|---|
| test_error | 0.20031 |
Threshold
Threshold = Median
| Measure | Value |
|---|---|
| threshold | 0.8627 |
| train_error | 0.1464 |
| test_error | 0.19025 |
Threshold = Median - Standard Deviation:
| Measure | Value |
|---|---|
| threshold | 0.645289077544 |
| train_error | 0.00452361737533 |
| test_error | 0.13335 |