基于正则项优化的高光谱图像噪声消除方法

Research on Image Noise Elimination Based on Regularization Term Optimization

为了有效消除多光谱和高光谱遥感图像中的重要噪声源(泊松分布噪声),提出了一种基于正则项优化的高光谱图像去噪方法。该方法根据光谱图像立方体矢量模型以及泊松分布噪声分析,利用一个全变分正则项对单色辐射图像的方法进行扩展,从而保留边缘信息并在光谱和空间上具有自适应性。此外,为了降低该方法的计算复杂度,采用基于分裂Bregman的方法进行了优化。实验结果显示,相比于最近的几种类似方法,该方法在合成和真实数据中均表现出较好的泊松分布噪声去除性能。

In order to eliminate the important noise sources(Poisson distribution noise)in multispectral andhyperspectral remote sensing images effectively,a hyperspectral image denoising method based on regulariza-tion optimization was proposed.Based on the spectral image cube vector model and Poisson distribution noiseanalysis,a full variational regularization term was used to extend the method of monochromatic radiation im-age,so as to preserve the edge information and be adaptive in the spectrum and space.In addition,in order toreduce the computational complexity of the proposed method,an optimization method based on split Bregmanwas adopted.The experimental results showed that compared with the most recent similar methods,the pro-posed method showed better removal performance of Poisson distribution noise in both synthetic and real data.