Expression Recognition Based on Low Pixel Face Images

Fu Liu; Maojun Li; Jianwen Hu; Yuhe Xiao; Zhan Qi

doi:10.3788/LOP57.101008

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 10 >Page 101008 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 10, 101008 (2020)

Expression Recognition Based on Low Pixel Face Images

Fu Liu, Maojun Li^*, Jianwen Hu, Yuhe Xiao, and Zhan Qi

Author Affiliations

School of Electrical and Information Engineering, Changsha University of Science and Technology, Changsha, Hunan 410114, China

show less

DOI: 10.3788/LOP57.101008 Cite this Article Set citation alerts

Fu Liu, Maojun Li, Jianwen Hu, Yuhe Xiao, Zhan Qi. Expression Recognition Based on Low Pixel Face Images[J]. Laser & Optoelectronics Progress, 2020, 57(10): 101008 Copy Citation Text

show less

Fig. 1. Pipeline of the facial expression preprocessing

Download full size

Fig. 2. CNN model

Download full size

Fig. 3. 8 kinds of facial expression images

Download full size

Fig. 4. Diagram of SoftMax average voting

Download full size

Layer	Type	Kernel size /pixel×pixel	Stride	Output number	Output size /pixel×pixel
Layer 1	Convolution	3×3	1	64	30×30
Layer 2	Convolution	3×3	1	128	28×28
Layer 3	Max pooling	2×2	2	128	14×14
Layer 4	Convolution	3×3	1	128	14×14
Layer 5	Convolution	3×3	1	128	14×14
Layer 6	Convolution	3×3	1	256	12×12
Layer 7	Max pooling	2×2	2	256	6×6
Layer 8	Convolution	3×3	1	256	6×6
Layer 9	Convolution	3×3	1	256	6×6
Layer 10	Average pooling	3×3	3	256	2×2
Layer 11	Convolution	2×2	1	512	1×1
Layer 12	Fully connected	-	-	-	160×1
Output	SoftMax	-	-	-	8×1

Table 1. Convolutional network parameters

Category	Angry	Disgust	Fearful	Happy	Sad	Surprised	Scorn	Neutral
Train database	14040	12220	7800	14300	8970	8580	5460	8970
Test database	27	24	15	28	15	16	12	15

Table 2. Facial expression dataset

Model	Originaldatabase	Local binarypatterndatabase	Histogramequalizationdatabase
Accuracy /%	89.5	86.9	90.5
Time/s	1.54	1.27	1.56

Table 3. Comparison of recognition accuracy and time

Category Accuracy	Angry	Disgust	Fearful	Happy	Sad	Surprised	Scorn	Neutral	Total
First experiment	96.3	91.7	60.0	100.0	86.7	100.0	83.3	93.3	90.5
Second experiment	88.9	91.7	86.7	100.0	80.0	100.0	83.3	93.3	91.2
Third experiment	96.3	95.8	60.0	100.0	86.7	93.8	75.0	93.3	89.8
Total	93.8	93.1	68.9	100.0	84.5	97.9	80.5	93.3	90.5

Table 4. Improved recognition accuracy of CNN modelunit: %

Model	LeNet-5	Without decision level
Accuracy /%	74.6	87.9
Time /s	0.59	0.31

Table 5. Comparison of recognition accuracy and time between two models

Fu Liu, Maojun Li, Jianwen Hu, Yuhe Xiao, Zhan Qi. Expression Recognition Based on Low Pixel Face Images[J]. Laser & Optoelectronics Progress, 2020, 57(10): 101008

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information