Recognition Method of Inscription Graphics Based on Dual Features of Topology and Mesh

Table 1. Feature dimension of different feature extraction algorithms
View table
Table 1. Feature dimension of different feature extraction algorithms
Class Propoesd algorithm SIFT HOG RestNet18
Feature structure $7 + 4 \times 2$ $128 \times n$ $36 \times 15 \times 15$ $7 \times 7 \times 512$
1 15 11136 8100 25088
2 15 25344 8100 25088
3 15 38912 8100 25088
4 15 66176 8100 25088
5 15 12672 8100 25088

Table 2. Feature extraction algorithm model efficiency
View table
Table 2. Feature extraction algorithm model efficiency
Classification model Algorithm Average running time /s
Bagging Propoesd algorithm 0.02
SIFT 1.53
HOG 1.31
RestNet18 RestNet18 0.68

Table 3. Accuracy of four feature extraction algorithmsunit: %
View table
Table 3. Accuracy of four feature extraction algorithmsunit: %
Five-fold
cross validation
Bagging RestNet18
T-MF SIFT HOG
Average 87.95 72.41 66.25 56.87
1 89.29 75.45 66.52 56.25
2 86.16 68.75 64.73 54.46
3 87.50 73.66 68.75 61.60
4 91.07 76.34 67.86 59.82
5 85.71 67.86 63.39 52.23

Tools

Get Citation

Copy Citation Text

Wenteng Liu, Huiqin Wang, Ke Wang, Zhan Wang. Recognition Method of Inscription Graphics Based on Dual Features of Topology and Mesh[J]. Laser & Optoelectronics Progress, 2022, 59(4): 0410018

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites