Motivation
| Original image |  |
| Binarized image with guides |  |
| Normalized image |  |
Figure 1. Example image camera-captured text and its normalized version. (Copyright 2014 IEEE. Used by permission.)
Many character recognition systems treat out-of-plane character orientation as "noise," which classifiers must learn to cope with. To enhance the discriminative power of our recognition system, we strive to normalize text regions before recognition by identifying the primary guide lines and warping the image accordingly.