A Hybrid Framework for Text Recognition Used in Commodity Futures Document Verification

The document verification is significantly important for regulatory agencies to supervise the commodity futures trading market. The intelligent recognition of unstructured text data can remarkably improve the efficiency of document verification. Traditional text recognition methods have some limitations for documents with fixed templates, such as lacking the ability to make full use of known information, which results in low detection efficiency. Meanwhile, the background of documents without stable templates is relatively complicated, and the text position is not fixed, which makes it difficult for existing methods to capture all text information through location information. In response to the above problems, this paper proposes a new hybrid framework for text recognition used in commodity futures document verification. In the text detection stage, manually labeled anchor points are used to determine the position of the valid text refer to the known template. The position of the text without known template is detected based on the end-to-end scene text recognition model. In the text recognition stage, Connectionist Temporal Classification (CTC) loss in speech recognition is introduced to realize the recognition of random length text. The hybrid framework proposed in this paper is applied to the text recognition of quality inspection certificates and business contracts, achieving reliable text recognition results and saving labor costs significantly.