研究目的
To develop a low-rank Bayesian tensor factorization approach for removing additive Gaussian noise from hyperspectral images by leveraging global correlation along the spectrum and nonlocal self-similarity across space.
研究成果
The proposed LBTF-HSI method effectively denoises hyperspectral images by integrating nonlocal self-similarity and global spectral correlation through Bayesian tensor factorization, achieving superior performance over state-of-the-art methods in both simulated and real scenarios, with automatic noise level adaptation and rank determination.
研究不足
The noise model assumes Gaussian distribution, which may not capture real-world noise complexities; computational efficiency is lower compared to some methods; parameters like patch size and number of iterations require tuning.
1：Experimental Design and Method Selection:
The method involves an iterative denoising framework with grouping, low-rank tensor recovery using Bayesian CP factorization, and aggregation steps. It uses variational Bayesian inference for model solving.
2：Sample Selection and Data Sources:
Experiments are conducted on simulated data from the Columbia HSI Dataset and real data from HYDICE urban dataset and Harvard real-world hyperspectral datasets.
3：List of Experimental Equipment and Materials:
No specific equipment or materials are mentioned; the method is implemented in software (Matlab2016a) on a computer with Intel Core i7-7700K CPU and 16 GB RAM.
4：Experimental Procedures and Operational Workflow:
The noisy HSI is split into overlapping full-band patches, grouped using block matching, denoised using the LBTF algorithm, and aggregated with iterative regularization.
5：Data Analysis Methods:
Performance is evaluated using quantitative metrics (PSNR, SSIM, FSIM, ERGAS, SAM) and visual comparisons.
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
