A Passport Text Information Recognition System Based on Deep Learning
It's meaningful to apply existing scene text detection and recognition algorithms based on deep learning to passports so as to implement structured output of the key information in multinational passport pictures.In this paper,Aiming at solving the problem of detecting texts with extreme aspect ratio and relatively small size,we use pixel segmentation based algorithm and conduct multi-scale feature fusion;to alleviate the interference in character pixel,we use recurrent neural network to model the context of picture features,so as to reduce the interference of defacing;to avoid the interference of irrelevant text,the multi-modal Transformer based on text information and layout information is used for modeling,so as to obtain multi-modal mode of key information,filtering out irrelevant information,matching and extracting key words.The experimental results show that the system has a lot of advantages.
passportdeep learningtext detectiontext recognitionkey information extraction
万方数据
维普
