Font information fusion-based medicine-taking bill recognition result error correction method

Abstract

The invention relates to a font information fusion-based medicine-taking bill recognition result error correction method and belongs to the field of character recognition. The method comprises the following steps of: constructing a standard medicine word bank, storing each piece of medicine information in the word bank in a BK tree memory structure as a node, setting a search distance threshold n, reducing a data search scale through a threshold search rule, and obtaining a result candidate set; carrying out similarity matching on a character string to be corrected after character recognition and a character string in a result candidate set, improving a traditional editing distance formula on the basis of an original similarity matching scheme, keeping the insertion and deletion operation cost unchanged, and reducing character replacement cost; during character replacement operation, considering relevant information of three fonts including five-stroke codes, four-corner codes and strokes, and improving character string approximate matching precision; and replacing the character string with the highest similarity as an error correction result. According to the method of the invention, a medicine-taking bill identification result is corrected, so that the medicine-taking bill identification accuracy is improved.

Description

technical field [0001] The invention belongs to the technical field of character recognition, and relates to a method for correcting errors in recognition results of medication lists based on fused font information. Background technique [0002] With the continuous improvement of people's living standards, the number of patients with chronic diseases is increasing year by year, and patients need long-term or even life-long medication to control and alleviate the disease. Long-term tracking of patients' medication information can help medical staff understand patients' actual medication conditions, and facilitate their health assessment and disease prediction. As an important basis for patient medication information, the medical medication list mainly includes doctor's order time, drug name, dosage and frequency information. However, it is not easy to save paper medication lists, manual data collection is labor-intensive, and there is a risk of data errors. OCR technology pr...

AI Technical Summary

Problems solved by technology

However, drug names are mainly spliced ​​by some medical-specific vocabulary and do not have a semantic environment. Semantic-based methods cannot be effectively combined with practical problems.

The edit distance algorithm measures the similarity between two strings by the minimum number of edit operations required to convert one into the other. It is mainly used in the similarity calculation of English strings, and the edit operation information considered is too Single, resulting in the traditional edit distance not meeting the requirements of actual application scenarios

Images