OCR text error correction method

The invention discloses an OCR (Optical Character Recognition) text error correction method. According to the method, a Dense Layer is linked behind an encoder to serve as a decoder. In the decoding process of the decoder, a search path of a solution space is optimized by using a beam search method, and candidate results ranked top K (K is the search width of beam search) are obtained. Probability distribution of candidate results is calculated through a logsoftmax function, and finally a candidate with the highest probability score is selected from the probability distribution to serve as an error-corrected text. Compared with the prior art that only pure text is used for OCR text error correction, the method has the advantages that the layout and visual information of the document can be better utilized, the error correction result is better improved, and the error correction performance on the data set SROIE is improved by about 20% compared with that of a pure text error correction scheme. 该发明公开了一种OCR文本纠错