Deep Learning based Text Detection Method for Personal Information De-identification

摘要

In order to de-identify personal information exposed as text in images obtained from CCTV, cameras, etc., a method for performing high-level text detection based on deep learning is provided. A text detection method according to an embodiment of the present invention includes generating a plurality of input images by augmenting an original input image; predicting text candidate regions by inputting the original input image and the input images generated in the generating step into an artificial intelligence model; merging predicted text candidate regions; correcting the positions of the merged text candidate regions; and deleting some of the text candidate regions whose positions have been corrected. Accordingly, through high-level text detection based on deep learning, it is possible to effectively detect and de-identify personal information exposed as text in real life scenes obtained from CCTV, cameras, etc.