Open AccessNecklace: a novel long text detection model
Author(s) -
Qin Xiao,
Chu Xutao,
Yuan Changan,
Zhao Qi
Publication year - 2020
Publication title -
the journal of engineering
Language(s) - English
Resource type - Journals
ISSN - 2051-3305
DOI - 10.1049/joe.2019.1176
Subject(s) - text detection , computer science , necklace , frame (networking) , artificial intelligence , feature (linguistics) , channel (broadcasting) , pattern recognition (psychology) , regression , line (geometry) , image (mathematics) , architecture , mathematics , statistics , geometry , telecommunications , combinatorics , art , linguistics , philosophy , visual arts
This study proposes a new text detection model and a novel tag generation method for long text detection in natural scenes. The background suppression module is added to the network architecture to integrate feature mapping of each channel, weaken background features, enhance text features, and improve detection accuracy. In the design of the regression frame, the use of elliptical geometric regression, through the formation of long text through a plurality of elliptical connections, better fit the shape of the text line to solve the problem of long text detection. The proposed method has achieved the performance of state‐of‐art in the popular datasets.