檜垣, 泰彦

須貝, 康雄

HIGAKI, Yasuhiko

SUGAI, Yasuo

ディープラーニング

自動マスキング

テキスト検出

CTPN

Plant specimen image

Deep learning

Automatic masking

Text detection

CTPN

[ABSTRACT] When plant specimens are published, the locality of endangered species cannot be disclosed. Images with specimen labels describing the collection points must be covered. So far, the text detection network CTPN has been used to detect the text on the sample and try to identify the location of the label, but due to false detection and the influence of characters other than the label, sufficient results have not been obtained. In low-resolution sample photos, it is difficult to directly detect the target to be masked using text recognition after text detection. Therefore, in this study, after performing text detection processing on the sample image, deleting the error detection and arranging the text box according to the characteristics of the plant specimen image, the method of identifying the target to be covered from the arrangement pattern was discussed. The results show that the method is effective.