Authors: Yang Qiu
Description: I first make a simple model based on VGG16 to split the raw data to Chinese part and English part. For English part, I combine the results from three models, ASTER, MORAN and SAR with novel post processing method. For Chinese part, I first cropped images with digit image processing methods aiming to have one character in each image, then do the character-level recognition based on the cropped images (here I make several cropped images datasets with different cropped methods), then I do another post processing method to combine the results based on different cropped datasets. Eventually I combine the English recognition results and Chinese recognition results to get the final results.