Efficient local stereo matching algorithm based on fast gradient domain guided image filtering
作者：
Highlights：
• By incorporating a multi-scale edge-aware weighting and sub-sampling strategy into GIF, a novel fast gradient domain guided image filter (F-GDGIF) is proposed, which achieves better edge-aware performance with a faster execution time.
• The proposed F-GDGIF is employed in cost aggregation. Due to its advantages of edge preserved smoothing property and low computational complexity, F-GDGIF based cost aggregation method achieves better performance compared to other GIF based methods.
• Apart from cost aggregation, F-GDGIF is also adopted in disparity refinement and achieved better results compared to GIF based method.
摘要
•By incorporating a multi-scale edge-aware weighting and sub-sampling strategy into GIF, a novel fast gradient domain guided image filter (F-GDGIF) is proposed, which achieves better edge-aware performance with a faster execution time.•The proposed F-GDGIF is employed in cost aggregation. Due to its advantages of edge preserved smoothing property and low computational complexity, F-GDGIF based cost aggregation method achieves better performance compared to other GIF based methods.•Apart from cost aggregation, F-GDGIF is also adopted in disparity refinement and achieved better results compared to GIF based method.
论文关键词：Stereo matching,Cost aggregation,Disparity refinement,Guided image filtering
论文评审过程：Received 23 April 2020, Revised 31 December 2020, Accepted 6 April 2021, Available online 18 April 2021, Version of Record 18 April 2021.
论文官网地址：https://doi.org/10.1016/j.image.2021.116280
