Laser & Optoelectronics Progress, Volume. 56, Issue 10, 101502(2019)
Stereo-Matching Algorithm Using Weighted Guided Image Filtering Based on Laplacian of Gaussian Operator
Figures & Tables(8)
Fig. 1. Comparison of different matching costs. (a) Single channel AD; (b) mean AD; (c) weighted AD
Fig. 2. Comparison of disparity maps before and after improvement by guided image filtering. (a) Before improvement; (b) after improvement
Fig. 3. Comparison of mismatch rates before and after improvement by guided image filtering
Fig. 4. Disparity selection and refinement process. (a) Matching result of left image; (b) test result of left-right consistency; (c) result of first interpolation; (d) result of second interpolation
Fig. 5. Comparison of details for two interpolations. (a) Result of first interpolation; (b) result of second interpolation
Fig. 6. Result of proposed algorithm. (a) Original left images; (b) ground truth; (c) disparity maps obtained by proposed algorithm; (d) mismatching points gained by proposed algorithm
Table 1. Mismatching rates of different stereo-matching algorithms in non-occluded regions and all regions of four groups of test image sets%
View tableTable 1. Mismatching rates of different stereo-matching algorithms in non-occluded regions and all regions of four groups of test image sets%
Algorithm Tsukuba Venus Teddy Cones Average Nonocc All Nonocc All Nonocc All Nonocc All GC-occ 1.19 2.01 1.64 2.19 11.20 17.40 5.36 12.40 6.67 RTCensus 5.08 6.25 1.58 2.42 7.96 13.80 4.10 9.54 6.34 Fast Aggreg 1.16 2.11 4.03 4.75 9.04 15.20 5.37 12.60 6.78 Semi Glob 3.26 3.96 1.00 1.57 6.02 12.20 3.06 9.75 5.10 Adapt Weight 1.38 1.85 0.71 1.19 7.88 13.30 3.97 9.79 5.01 RTAdapt Weight 1.45 1.99 0.40 0.81 7.65 13.30 3.48 9.34 4.80 Proposed algorithm 1.81 2.65 0.27 0.51 6.75 10.15 2.78 9.62 4.32
Table 2. Mismatching rates of different stereo-matching algorithms in non-occluded regions of other 27 groups of test image sets%
View tableTable 2. Mismatching rates of different stereo-matching algorithms in non-occluded regions of other 27 groups of test image sets%
Stereo Dataset Adapt Weight MST Fast Cost Proposed algorithm Aloe 6.502 6.613 7.064 5.161 Art 12.883 13.744 12.032 8.961 Baby1 5.453 7.654 3.161 3.602 Baby2 13.643 14.124 3.971 4.252 Baby3 8.763 9.534 4.711 6.092 Books 12.653 12.974 10.212 8.681 Bowling1 16.873 19.144 6.211 11.652 Bowling2 10.433 12.634 8.182 7.661 Cloth1 1.253 1.182 1.884 0.891 Cloth2 5.564 5.403 4.202 3.071 Cloth3 3.814 3.203 2.792 1.921 Cloth4 3.214 2.533 2.252 1.621 Dolls 6.432 7.864 7.123 4.511 Flowerpots 15.443 17.274 9.631 13.612 Lampshade1 12.364 11.292 12.123 10.801 Lampshade2 17.443 14.202 12.481 22.594 Laundry 16.892 19.393 20.644 13.531 Midd1 36.472 18.961 36.953 40.364 Midd2 35.563 20.211 34.022 38.084 Moebius 13.264 11.333 10.642 9.741 Monopoly 15.431 16.852 19.743 24.064 Plastic 32.443 30.302 27.061 41.094 Reindeer 11.733 11.964 7.842 4.941 Rocks1 5.434 4.143 3.972 2.241 Rocks2 5.024 3.333 2.672 1.391 Wood1 8.993 11.174 4.932 3.981 Wood2 6.154 4.203 2.712 0.481 Average 12.594 11.523 10.341 10.922
Tools
Get Citation
Copy Citation Text
Bo Zhou, Ling Qin, Wei Gong. Stereo-Matching Algorithm Using Weighted Guided Image Filtering Based on Laplacian of Gaussian Operator[J]. Laser & Optoelectronics Progress, 2019, 56(10): 101502
Paper Information
Category: Machine Vision
Received: Nov. 27, 2018
Accepted: Dec. 14, 2018
Published Online: Jul. 4, 2019
The Author Email: Zhou Bo (zhoubo180@foxmail.com), Qin Ling (qinling@whut.edu.cn), Gong Wei (gongwei_whut@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20