研究目的
To refine the depth map generated by the Adaptive Random Walk with Restart (ARWR) algorithm in order to obtain significant improvements in accuracy.
研究成果
The proposed post-processing technique increases the accuracy of the depth map computed by ARWR method, keeping the sharp edges and corners along with the main structure of the reference image. The method was proven effective through comparison with top methods from the Middlebury benchmark and Google’s stereo matching method.
研究不足
The processing time is a trade-off, being poor but can be improved for real-time applications with low processing power.
1：Experimental Design and Method Selection:
The framework starts with the ARWR algorithm, followed by a median solver/filter based on mutual structure, enhanced by a joint filter, and a transformation in image domain to remove artifacts.
2：Sample Selection and Data Sources:
Standard images from the Middlebury ‘dense’ training dataset were used.
3：List of Experimental Equipment and Materials:
Not explicitly mentioned.
4：Experimental Procedures and Operational Workflow:
The process involves extracting the initial depth using ARWR, applying mutual joint weighted median filter, overwriting the structure of the RGB image on the depth map, and transferring the depth map to a signal for normalized interpolated convolution.
5：Data Analysis Methods:
The proposed method was compared with the top eight algorithms in the Middlebury benchmark and Google’s depth map estimation method using metrics like MSE, PSNR, SNR, SSIM, and DSSIM.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
