基于图像分割的稠密立体匹配算法 下载: 1087次
Dense Stereo Matching Algorithm Based on Image Segmentation
1 中国科学院沈阳自动化研究所, 辽宁 沈阳 110016
2 东北大学信息科学与工程学院, 辽宁 沈阳 110819
3 中国科学院光电信息处理重点实验室, 辽宁 沈阳 110016
图 & 表
图 1. 算法流程图。(a)匹配代价聚合;(b)视差后处理;(c)综合流程图
Fig. 1. Flow chart of algorithm. (a) Matching cost aggregation; (b) disparity post-processing; (c) integrated flow chart
下载图片 查看原文
图 2. SLIC算法结果。(a) Tsukuba图像;(b)超像素分割图;(c)边缘图像
Fig. 2. Result of SLIC algorithm. (a) Tsukuba image; (b) superpixel segmentation image; (c) edge image
下载图片 查看原文
图 3. 不同算法所得结果。(a)孔洞填充;(b)十字交叉自适应窗口加权中值滤波
Fig. 3. Results of different algorithms. (a) Hole filling; (b) cross adaptive window weighted median filtering
下载图片 查看原文
图 4. 支持域
Fig. 4. Support region
下载图片 查看原文
图 5. 不同参数对平均误差的影响。(a) N;(b) β和γ;(c) ρ;(d) T
Fig. 5. Influences of parameters on AvgPBM. (a) N; (b) β and γ; (c) ρ; (d) T
下载图片 查看原文
图 6. 实验结果。(a)左图像;(b)真实视差图;(c)所提算法结果;(d) MST算法结果;(e) GF算法结果;(f) GA-DP算法结果
Fig. 6. Experimental results. (a) Left image; (b) ground-truth disparity; (c) result of proposed algorithm; (d) result of MST algorithm; (e) result of GF algorithm; (f) result of GA-DP algorithm
下载图片 查看原文
表 1不同算法下的平均误匹配率
Table1. AvgPBM for different algorithms%
Algorithm | Tsukuba | | Venus | | Teddy | | Cones | Avg PBM | Avg Disc |
---|
n-occ | all | disc | n-occ | all | disc | n-occ | all | disc | n-occ | all | disc |
---|
Proposed | 1.50 | 1.95 | 6.71 | | 0.11 | 0.33 | 1.25 | | 5.27 | 10.8 | 14.5 | | 2.38 | 8.02 | 7.01 | 4.99 | 7.36 | MST | 1.47 | 1.85 | 7.88 | | 0.25 | 0.42 | 2.60 | | 6.01 | 11.6 | 14.3 | | 2.87 | 8.45 | 8.10 | 5.48 | 8.08 | GF | 1.51 | 1.85 | 7.61 | | 0.20 | 0.93 | 2.42 | | 6.16 | 11.8 | 16.0 | | 2.71 | 8.24 | 7.66 | 5.55 | 8.42 | GA-DP | 1.57 | 2.00 | 7.32 | | 0.89 | 1.00 | 3.18 | | 7.20 | 12.4 | 16.1 | | 3.68 | 9.18 | 8.62 | 6.10 | 8.80 | Gray | 1.91 | 2.74 | 9.70 | | 0.32 | 0.68 | 4.25 | | 5.99 | 11.7 | 16.2 | | 3.70 | 9.64 | 10.7 | 6.46 | 10.21 |
|
查看原文
表 2不同算法下n-occ的平均误匹配率
Table2. AvgPBM for different algorithms on n-occ region%
Stereo pair | Proposed | GF | CS-MST | Gray | MST |
---|
Tsukuba | 1.502 | 1.513 | 2.125 | 1.914 | 1.471 | Venus | 0.111 | 0.202 | 0.845 | 0.324 | 0.253 | Teddy | 5.271 | 6.165 | 7.614 | 5.993 | 5.532 | Cones | 2.381 | 2.712 | 4.104 | 3.703 | 6.015 | Alone | 4.582 | 5.534 | 4.141 | 7.135 | 4.633 | Art | 7.081 | 9.032 | 9.793 | 9.884 | 10.795 | Baby1 | 2.621 | 4.693 | 7.374 | 3.242 | 8.395 | Baby2 | 3.301 | 6.083 | 11.954 | 4.912 | 13.375 | Baby3 | 3.461 | 5.794 | 5.643 | 4.522 | 7.255 | Books | 8.291 | 10.223 | 9.562 | 10.645 | 10.264 | Bowling1 | 6.481 | 14.523 | 16.814 | 9.772 | 20.895 | Bowling2 | 4.871 | 7.083 | 9.314 | 6.822 | 10.155 | Cloth1 | 1.013 | 1.084 | 0.511 | 1.125 | 0.612 | Cloth2 | 2.311 | 3.463 | 2.852 | 3.574 | 4.135 | Cloth3 | 1.461 | 2.153 | 1.772 | 2.204 | 2.665 | Cloth4 | 3.204 | 1.622 | 1.301 | 3.745 | 1.873 | Dolls | 4.081 | 5.043 | 5.002 | 6.575 | 5.954 | Flowerpots | 9.801 | 12.792 | 16.674 | 12.883 | 19.415 | Lampshade1 | 5.591 | 11.574 | 10.433 | 6.742 | 11.995 | Lampshade2 | 13.881 | 21.135 | 20.884 | 15.042 | 18.203 | Laundry | 15.653 | 16.404 | 13.692 | 18.505 | 12.941 | Midd1 | 40.104 | 40.115 | 32.322 | 36.673 | 27.851 | Midd2 | 39.245 | 35.853 | 34.502 | 36.934 | 32.091 | Moebius | 7.441 | 9.254 | 7.672 | 9.335 | 8.693 | Monopoly | 25.403 | 27.995 | 22.511 | 27.714 | 24.212 | Plastic | 33.621 | 39.292 | 42.534 | 40.213 | 47.035 | Reindeer | 27.574 | 7.231 | 9.152 | 28.365 | 9.873 | Rocks1 | 3.684 | 2.702 | 2.231 | 4.415 | 2.833 | Rocks2 | 2.043 | 1.612 | 1.571 | 2.585 | 2.084 | Wood1 | 3.772 | 2.831 | 8.684 | 4.733 | 11.065 | Wood2 | 2.272 | 2.343 | 0.991 | 3.004 | 5.615 | AvgErr | 9.42 | 10.25 | 10.44 | 10.74 | 11.23 | AvgRank | 1.87 | 3.06 | 2.61 | 3.67 | 3.65 |
|
查看原文
马瑞浩, 朱枫, 吴清潇, 鲁荣荣, 魏景阳. 基于图像分割的稠密立体匹配算法[J]. 光学学报, 2019, 39(3): 0315001. Ruihao Ma, Feng Zhu, Qingxiao Wu, Rongrong Lu, Jingyang Wei. Dense Stereo Matching Algorithm Based on Image Segmentation[J]. Acta Optica Sinica, 2019, 39(3): 0315001.