Innovative Text Extraction Algorithm Based on TensorFlow

Abstract

Extracting business registration information exploiting graphic recognition algorithms on the Internet nowadays is vital to e-commercial business. However, business registration information is usually presented in graphics and existing graphic recognition systems have been hindered because of their slow detection speed, low accuracy, and complex operations. Thereby, we propose an innovative text extraction algorithm based on TensorFlow (TEAT). We first utilize the web crawler to obtain the data source, and then extract the character information by using our TEAT based on TensorFlow framework recognition technology. Our TEAT algorithm can extract business registration information efficiently and effectively. Comparing with existing text extraction algorithm based on Tess4j framework for extracting Tmall shop business license picture information, our TEAT has obvious advantages over Tess4j framework with higher accuracy and efficiency.