Title | ||
---|---|---|
Inscription Segmentation Using Synthetic Inscription Images for Text Detection at Stone Monuments |
Abstract | ||
---|---|---|
Stone monuments have historical value, and the inscriptions engraved on them can tell us about the events and people at the time of their installation. Photography is an easy way to record inscriptions; however, the light falling on the monument, the resulting shadows, and the innate texture of the stone can make the text in the photographs unclear and difficult to recognize. This paper presents a method for inferring pixel-wise text areas in a stone monument image by developing a deep learning network that can deduce the shape of kanji characters. Our method uses pseudo-inscription images for training a deep neural network, which is generated by synthesizing a shaded image representing the engraved text and stone texture image. Through experiments using a High Resolution Net (HRNet), we confirm that the HRNet achieves high accuracy in the task of inscription segmentation and that training with pseudo-inscription images is effective in detecting inscriptions on real stone monuments. Thus, synthetic inscription images can facilitate efficient and accurate detection of text on stone monuments, thereby contributing to further history research. |
Year | DOI | Venue |
---|---|---|
2021 | 10.1007/978-3-030-86198-8_13 | DOCUMENT ANALYSIS AND RECOGNITION, ICDAR 2021 WORKSHOPS, PT I |
Keywords | DocType | Volume |
Synthetic data, Inscription, Text detection, Deep learning | Conference | 12916 |
ISSN | Citations | PageRank |
0302-9743 | 0 | 0.34 |
References | Authors | |
0 | 7 |
Name | Order | Citations | PageRank |
---|---|---|---|
Naoto Morita | 1 | 0 | 0.34 |
Ryunosuke Inoue | 2 | 0 | 0.34 |
Masashi Yamashiro | 3 | 0 | 1.01 |
Takatoshi Naka | 4 | 0 | 1.01 |
Atsuko Kanematsu | 5 | 0 | 0.34 |
Shinya Miyazaki | 6 | 53 | 12.33 |
Jun-ichi Hasegawa | 7 | 221 | 61.17 |