Recognition of Digits on License Plate by RAISR with Changing Contrast Ratio

We have researched unreadable digits on license plates of cars could be read by machine learning classifiers trained with CG images. Digits on license plates can be evidences for climes if they can be read and CG does not require huge numbers of real images for training. We reached a conclusion that some digits could be read with high accuracy when training images were effectively extended. We also found that test images with emphasized edges were effectively enhanced by image resolution enhancement. Therefore, in this paper, we evaluated how much edge emphasis by changing contrast ratio influences the resolution enhancement. The results show that enhancement of contrast ratio increases classification accuracy of unreadable digits with the resolution enhancement. On the other hand, the contrast ratio enhancement does not work for images with too low contrast ratio. Also, it decreases the classification accuracy on the contrary when the contrast ratio of original images is enough high. We evaluated which contrast ratio suits edge emphasis.