Laser & Optoelectronics Progress, Volume. 56, Issue 1, 011008(2019)
Convolution-Deconvolution Image Segmentation Model for Fusion Features and Decision
Figures & Tables(10)
Fig. 1. Convolution-deconvolution image segmentation model for fusion features and decision
Fig. 2. Flow charts of data processing. (a) Data processing of CD-FFD; (b) each branch network data processing of CD-FFD
Fig. 3. Segmentation results of CD-FFD model. (a) RGB image; (b) gray image; (c) segmentation result of RGB-Net; (d) segmentation result of GRAY-Net; (e) segmentation result of CD-FFD; (f) ground-truth
Fig. 4. Segmentation results of CD-FFD model. (a) IRRG image; (b) DSM image; (c) segmentation result of IRRG-Net; (d) segmentation result of DSM-Net; (e) segmentation result of CD-FFD; (f) ground-truth
Fig. 5. Segmentation results of other images by CD-FFD model. (a) Original RGB images; (b) segmentationresults of CD-FFD
Table 1. Evaluation results of CD-FFD on 200 validation images from Weizmann Horse
View tableTable 1. Evaluation results of CD-FFD on 200 validation images from Weizmann Horse
Type Average COM Average global acc Average IOU RGB-Net 0.9183 0.9542 0.7815 GRAY-Net 0.9169 0.9407 0.7327 CD-FFD 0.9329 0.9656 0.8188
Table 2. Evaluation results of 128 test images from Weizmann horse dataset
View tableTable 2. Evaluation results of 128 test images from Weizmann horse dataset
Type Average COM Average global acc Average IOU RGB-Net 0.9279 0.9672 0.8834 GRAY-Net 0.9360 0.9648 0.8753 CD-FFD 0.9407 0.9708 0.8949
Table 3. Evaluation results of 5 test images from vaihigen dataset
View tableTable 3. Evaluation results of 5 test images from vaihigen dataset
Type Average COM Average global acc Average IOU IRRG-Net 0.9316 0.9668 0.8698 DSM-Net 0.9221 0.9610 0.8420 CD-FFD 0.9496 0.9748 0.8966
Table 4. Comparison among existing results of 128 test images from Weizmann Horse dataset
View tableTable 4. Comparison among existing results of 128 test images from Weizmann Horse dataset
Number Method Average global acc Average IOU 1 Ref. [16] 94.6 80.1 2 Ref. [17] 95.8 84.0 3 Ref. [18] 95.7 84.0 4 Ref. [19] 94.9 79.9 5 CD-FFD 97.2 89.5 6 CD-FFD+CRF 97.6 90.1
Table 5. Comparison among existing results of 5 test images from vaihigen dataset
View tableTable 5. Comparison among existing results of 5 test images from vaihigen dataset
Number Method Average global acc 1 SegNet[ 20 ]0.9078 2 CNN+RF[ 21 ]0.9423 3 CNN+RF+CRF[ 21 ]0.9430 4 Ref. [22] 0.9450 5 CD-FFD 0.9748
Tools
Get Citation
Copy Citation Text
Chenxiao Feng, Xili Wang. Convolution-Deconvolution Image Segmentation Model for Fusion Features and Decision[J]. Laser & Optoelectronics Progress, 2019, 56(1): 011008
Paper Information
Category: Image Processing
Received: Aug. 8, 2018
Accepted: Sep. 18, 2018
Published Online: Aug. 1, 2019
The Author Email: Wang Xili (wangxili@snnu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20