Laser & Optoelectronics Progress, Volume. 55, Issue 11, 111010(2018)
Sign Language Semantic Recognition Based on Optimized Fully Convolutional Neural Network
Figures & Tables(8)
Fig. 4. Noised images. (a) Gaussian noise with variance of 0.01; (b) Gaussian noise with variance of 0.02; (c) Gaussian noise with variance of 0.03
Table 1. Parameters of fully convolutional neural network
View tableTable 1. Parameters of fully convolutional neural network
Parameter layer Kernal size / (pixel×pixel) Input / (pixel×pixel) C1 7×7 58×58@6 dS1 2×2 29×29@6 C2 6×6 24×24@8 dS2 2×2 12×12@8 C3 5×5 8×8@10 dS3 2×2 4×4@10 C4 3×3 2×2@12 uS1 2×2 4×4@12 uS2 2×2 8×8@12 uS3 8×8 64×64@12
Table 2. Gesture recognition rate by FCN algorithm%
View tableTable 2. Gesture recognition rate by FCN algorithm%
Item Gesture category S T U V W Recognition rate 96.92 96.85 97.86 97.93 97.50
Table 3. Recognition rate by FCN algorithm under different Gaussian noises%
View tableTable 3. Recognition rate by FCN algorithm under different Gaussian noises%
Gaussian noise variance Gesture category S T U V W 0.01 96.74 96.72 97.34 97.37 97.18 0.02 96.31 96.27 97.04 97.31 96.87 0.03 96.15 96.02 96.83 97.03 96.53
Table 4. Performance comparison between FCN and other algorithms%
View tableTable 4. Performance comparison between FCN and other algorithms%
Algorithm Gesture category S T U V W Algorithm in Ref. [4] 95.86 95.76 95.83 96.03 96.12 Algorithm in Ref. [6] 95.94 96.18 96.22 96.38 96.35 Algorithm in Ref. [8] 96.21 96.43 96.32 96.86 97.15 FCN algorithm 96.92 96.85 97.86 97.93 97.50
Tools
Get Citation
Copy Citation Text
Min Wang, Jing Hao, Chenhong Yao, Qiqi Shi. Sign Language Semantic Recognition Based on Optimized Fully Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2018, 55(11): 111010
Paper Information
Category: Image Processing
Received: May. 3, 2018
Accepted: Jun. 8, 2018
Published Online: Aug. 14, 2019
The Author Email: Jing Hao (haoanjing12@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20