Acta Optica Sinica, Volume. 37, Issue 11, 1115004(2017)
Anti-Noise Stereo Matching Algorithm Based on Improved Census Transform and Outlier Elimination
Figures & Tables(10)
Fig. 1. (a) Adaptive window with noise; (b) part of the Teddy map added with 0.05% salt and pepper noise; (c) disparity map computed by traditional cross-based cost aggregation algorithm; (d) disparity map computed by proposed cost aggregation algorithm
Fig. 2. (a) Arm length of pixel p in different directions; (b) aggregation region S of the pixel p
Fig. 3. Influence of α and β on error matching rate in non-occlusion region. (a) α values; (b) β values
Fig. 4. Error matching rate of different algorithms under different noises. (a) Salt and pepper noise; (b) Gaussian noise
Fig. 5. Experimental results of the benchmark images. (a) Middlebury benchmark images; (b) Middlebury benchmark disparity maps; (c) disparity maps with proposed algorithm; (d) error maps with proposed algorithm
Fig. 6. Experimental results of the benchmark images. (a) Middlebury benchmark images; (b) Middlebury benchmark disparity maps; (c) disparity maps with proposed algorithm
Table 1. Average error matching rate in non-occlusion region%
View tableTable 1. Average error matching rate in non-occlusion region%
Algorithm Noiseless 2% noise 5% noise 10% noise 15% noise Mei’s aggregation 4.54 6.67 11.73 26.88 48.63 Proposed 4.15 4.92 7.04 16.41 37.15
Table 2. Average error matching rate of different algorithms in non-occlusion region%
View tableTable 2. Average error matching rate of different algorithms in non-occlusion region%
Algorithm Noiseless Salt and pepper noise Gaussian noise 2% 5% 10% 15% Variance of 5 Variance of 10 Variance of 15 Variance of 20 CT 3.24 3.65 5.56 20.73 52.41 6.41 30.98 45.25 53.61 AverageCT 4.54 6.17 12.67 30.43 49.01 8.48 28.14 41.12 49.60 Min-CT 5.67 8.34 13.39 28.33 54.95 11.79 33.68 48.25 57.49 Proposed 3.57 3.98 5.14 10.48 24.57 6.98 28.87 40.87 48.94
Table 3. Average matching accuracy with different PSNR values in non-occlusion area%
View tableTable 3. Average matching accuracy with different PSNR values in non-occlusion area%
Algorithm Noiseless Salt and pepper noise Gaussian noise 50 dB 45 dB 40 dB 35 dB 30 dB 50 dB 45 dB 40 dB 35 dB 30 dB CT 96.76 96.55 96.02 94.44 47.59 19.88 95.58 93.79 91.27 87.57 54.75 AverageCT 95.46 94.24 91.88 87.33 50.99 21.77 94.48 92.72 89.55 83.13 58.88 Min-CT 94.33 93.11 91.17 86.61 45.05 15.91 91.11 88.87 85.63 77.11 41.75 Proposed 96.43 96.17 95.87 94.86 75.43 29.05 95.47 93.61 91.32 87.74 59.13
Table 4. Error matching rate of different algorithms when the threshold is 1%
View tableTable 4. Error matching rate of different algorithms when the threshold is 1%
Algorithm Tsukuba Venus Teddy Cones Avg n-occ all disc n-occ all disc n-occ all disc n-occ all disc Proposed 2.04 2.54 10.79 0.23 0.86 3.06 5.96 10.57 13.46 2.26 6.95 9.75 5.71 RandomVote 4.58 5.54 17.70 0.13 0.45 1.86 5.40 9.54 14.80 2.62 7.93 7.54 6.53 EnhencedBP 0.94 1.74 5.05 0.35 0.86 4.34 8.11 13.30 18.50 5.09 11.10 11.00 6.69 FBS 2.38 2.80 10.40 0.34 0.92 4.45 9.83 15.30 20.30 3.10 9.31 8.59 7.92 RTCensus 5.08 6.25 19.20 15.80 2.42 14.20 7.96 13.80 20.30 4.10 9.54 12.20 9.73 RinCensus 4.78 6.00 14.40 1.11 1.76 7.91 9.79 17.30 26.10 8.09 16.20 17.60 10.90
Tools
Get Citation
Copy Citation Text
Xinjun Peng, Jun Han, Yong Tang, Yuzhi Shang, Yujin Yu. Anti-Noise Stereo Matching Algorithm Based on Improved Census Transform and Outlier Elimination[J]. Acta Optica Sinica, 2017, 37(11): 1115004
Paper Information
Category: Machine Vision
Received: May. 31, 2017
Accepted: --
Published Online: Sep. 7, 2018
The Author Email: Xinjun Peng (yinizhishu@shu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20