Laser & Optoelectronics Progress, Volume. 58, Issue 20, 2010010(2021)
Color Constancy Algorithm of Microscopic Images Based on Autoencoder
Figures & Tables(11)
Fig. 1. Physical image of the microscope and camera. (a) Color temperature is 4500 K; (b) color temperature is 6300 K; (c) color temperature is 7000 K
Fig. 2. Statistics of the color cast coefficient. (a) Color temperature is 4500 K; (b) color temperature is 6300 K; (c) color temperature is 7000 K
Fig. 4. Color restoration results of different autoencoders. (a) Original image; (b) autoencoder; (c) UNet autoencoder
Fig. 7. Subjective results of different algorithms in microscope color constancy dataset. (a) RAW; (b) Ref.[8]; (c) Ref. [9]; (d) Ref. [10] ; (e) Ref. [11]; (f) Ref. [12]; (g) Ref. [13]; (h) Ref. [15] ; (i) ours; (j) GT
Table 1. Selection of threshold T
View tableTable 1. Selection of threshold T
T Best25% Mean Medium Trimean Worst25% 10 1.78 2.40 2.01 2.12 7.05 20 1.75 2.36 2.00 2.09 6.56 30 1.62 2.32 2.36 2.01 6.34 40 1.55 2.30 2.09 1.97 5.98 50 1.43 2.29 1.91 1.93 5.52 60 1.29 2.25 1.79 1.90 5.13 70 1.11 2.23 1.76 1.85 4.98 80 1.02 2.12 1.70 1.80 4.78 90 0.92 2.01 1.67 1.72 4.05 100 1.04 2.08 1.69 1.75 4.28 110 1.23 2.18 1.72 1.79 4.78 120 1.11 2.21 1.75 1.82 5.06 130 1.33 2.22 1.89 1.92 5.98 140 1.42 2.26 1.94 1.98 6.25 150 1.48 2.35 1.98 2.01 6.43 160 1.53 2.39 2.02 2.06 6.52 170 1.62 2.35 2.05 2.09 6.75 180 1.72 2.45 2.08 2.14 7.12
Table 2. Evaluation results of different algorithms in NUS-8 CC dataset
View tableTable 2. Evaluation results of different algorithms in NUS-8 CC dataset
Algorithm Best25% Mean Medium Trimean Worst25% GW[ 8 ]1.16 4.59 3.46 3.81 9.85 WP[ 9 ]1.44 9.91 7.44 8.78 21.27 Quasi-unsupervised[ 14 ]-- 1.97 1.91 -- -- CM[ 16 ]0.50 2.25 1.59 1.74 5.13 Ref.[15] 0.52 2.05 1.50 -- 4.48 Ref.[12] 0.46 2.18 1.48 1.64 5.03 Ref.[17] 0.50 2.39 1.61 1.74 5.67 Ref.[17]( pretrained) 0.46 2.35 1.55 1.73 5.62 CNN[ 13 ]0.68 2.14 1.83 1.95 4.26 UNet[ 19 ]1.13 2.56 2.04 2.23 5.17 Ours 0.92 2.01 1.67 1.72 4.05
Table 3. Evaluation results of different algorithms in RECommended CC dataset
View tableTable 3. Evaluation results of different algorithms in RECommended CC dataset
Algorithm Best25% Mean Medium Trimean Worst25% GW[ 8 ]5.00 9.70 10.00 10.00 13.70 WP[ 9 ]2.20 9.10 6.70 7.80 18.90 SoG[ 10 ]2.30 7.30 6.80 6.90 12.80 GE[ 11 ]0.70 5.50 3.30 3.90 13.80 CC-GANs (Pix2Pix)[ 18 ]1.20 3.60 2.80 3.10 7.20 CC-GANs(CycleGAN)[ 18 ]0.70 3.40 2.60 2.80 7.30 CC-GANs (StarGAN)[ 18 ]1.70 5.70 4.90 5.20 10.50 CNN[ 13 ]0.80 2.60 2.00 2.10 4.00 UNet[ 19 ]1.17 2.98 2.45 2.71 5.29 Ours 0.96 2.35 2.05 2.18 3.98
Table 4. Evaluation results of different algorithms in self-built microscope CC dataset
View tableTable 4. Evaluation results of different algorithms in self-built microscope CC dataset
Algorithm Best25% Mean Medium Trimean Worst25% GW[ 8 ]2.26 5.77 4.81 5.07 11.94 WP[ 9 ]4.43 6.19 5.04 5.21 10.60 SoG[ 10 ]1.01 3.59 3.17 3.18 7.51 GE[ 11 ]4.29 6.73 6.24 6.39 11.42 Ref.[12] 1.43 3.70 3.17 3.26 7.55 Ref.[15] 3.57 5.70 5.54 5.53 9.76 CNN[ 13 ]0.75 1.98 1.75 1.81 4.25 UNet[ 19 ]0.67 1.56 1.34 1.45 3.16 Ours 0.43 0.97 0.75 0.79 2.08
Tools
Get Citation
Copy Citation Text
Fangming Lan, Zongju Peng, Zhihua Lu, Qichao Shi, Fen Chen. Color Constancy Algorithm of Microscopic Images Based on Autoencoder[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2010010
Paper Information
Category: Image Processing
Received: Nov. 16, 2020
Accepted: Jan. 6, 2021
Published Online: Oct. 13, 2021
The Author Email: Peng Zongju (pengzongju@126.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20