Table 2 Methods for Arabic text detection in natural scenes.
Ref. | Algorithm | Dataset | Size | No. of images | Backbone | Text type | Evaluation | ||
---|---|---|---|---|---|---|---|---|---|
Precision (%) | Recall (%) | F score (%) | |||||||
CTPN | ASAYAR | 1920 × 1080 | 1375 | VGG | Horizontal text | 88 | 95 | 86 | |
EAST | 93 | 74 | 82 | ||||||
TextBoxes++ | 66 | 52 | 58 | ||||||
CTPN | ATIICA | N/A | 1180 | VGG | Horizontal text | 67 | 85 | 74.9 | |
EAST | 71 | 89 | 78.9 | ||||||
N/A | Private dataset | N/A | 575 | VGG | Multi-oriented dataset | 65.1 | 71.4 | 68 | |
Deep learning | TSVD | 640 × 640 | 7000 | VGG | Horizontal text | 89.9 | N/A | N/A | |
ICDAR2017 | 1000 | Multi-oriented and curved | 82 | ||||||
ICDAR2019 | 1000 | 82 | |||||||
Deep active learning | TSVD | 700 | Horizontal text | 83 | N/A | N/A | |||
ICDAR2017 | 250 | Multi-oriented and curved | 74 | ||||||
ICDAR2019 | 250 | 74 | |||||||
N/A | Private dataset | 1920 × 1080 | 50 | N/A | Horizontal text | 78 | 89 | 83 |