Authors: Gordon

Description: 1) preprocess images to locate main part of content.
2) use CTPN pretrained with MLT to finetune the data