研究目的
To propose an efficient super-resolution image reconstruction method for geometrically deformed remote sensing images, based on the nonlocal total variation (NLTV) regularization.
研究成果
The proposed method is able to produce HR images with better quality than the BTV based method. It is suggested to further reduce the computational cost and simplify the formulation in the near future.
研究不足
The computational cost for implementing NLTV is relatively high due to the calculation of similarity between two local patches which inhibits the feasibility of gradient descent type of algorithms.
1：Experimental Design and Method Selection:
The proposed method is based on NLTV regularization and L1-norm data fidelity to promote robustness and performance enhancement. A fast numerical algorithm is derived by taking advantage of dual formulations.
2：Sample Selection and Data Sources:
A sequence of 64 LR images of size 100×100 with intensity range [0,255] is generated by applying geometric deformation, removal of dark boundaries, convolution with a 3×3 Gaussian kernel, and downsampling with a sampling factor of 4 to the ground truth.
3：List of Experimental Equipment and Materials:
MATLAB R2016b on a desktop computer with 64GB RAM and a
4：2GHz Intel Xeon CPU E5-2650 vExperimental Procedures and Operational Workflow:
The proposed algorithm is compared with the fast robust SR method (BTV) on noise-free and noisy cases. Parameters are selected to optimize performance.
5：Data Analysis Methods:
Performance is quantified using the peak-to-noise ratio (PSNR).
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
