基于剪切波和全变分的农田遥感图像去噪去伪影方法

发布时间：2018-05-14 04:23

本文选题：图像处理 + 算法　；参考：《农业工程学报》2017年S1期

【摘要】：农田遥感图像在采集过程中会受到噪声影响,为得到准确的农田遥感图像数据,应对获取的农田遥感图像进行去噪预处理。农田遥感图像中的纹理承载了重要信息,在图像降噪的同时保持或增强图像纹理具有重要意义。由于纹理和噪声一样,在频域表现为高频信号,以分解和重构算法为基础的常见滤波(含小波变换)方法在降噪的同时,也会造成纹理清晰度的下降。该文结合农田遥感图像纹理呈现出来的直线特性,将剪切波(Shearlet)和变分理论相结合,提出了一种新的遥感农田图像保纹理降噪方法。该方法首先对较大的遥感图像分块进行shearlet变换,在降噪的同时识别不同图块图像的纹理含量;对细小纹理含量较少的平滑区域,采用保边降噪变分模型去除shearlet变换带来的人工伪影。为避免子图块边界带来的边界效应,该文基于中心仿射变换理论提出了一种新的图像延拓方法,有效提高了图像降噪的效果。试验结果表明,该文算法去噪后的峰值信噪比(peak signal to noise ratio,PSNR)平均值比全变分模型去噪算法大1 d B,该文算法去噪后的PSNR平均比曲线波去噪算法大2 d B。同基于Symmlet小波的Shearlet算法相比,该文算法处理后农田遥感图像中伪影减少,在高斯噪声标准偏差σ为10、20和30 d B时,峰值信噪比PSNR分别提高了13.99%、9.69%和7.75%。
[Abstract]:The field remote sensing image will be affected by noise in the process of acquisition. In order to obtain accurate farmland remote sensing image data, the field remote sensing image should be de-noised and preprocessed. The texture in the field remote sensing image bears important information, so it is of great significance to keep or enhance the texture of the image while reducing the noise of the image. Because texture and noise are represented as high frequency signals in frequency domain, the common filtering (including wavelet transform) method based on decomposition and reconstruction algorithm can reduce the noise and decrease the texture clarity at the same time. In this paper, a new texture preserving denoising method for remote sensing field images is proposed by combining the linear characteristics of remote sensing images with shear wave Shearlet (Shearlet) and variational theory. In this method, the larger remote sensing images are divided into blocks by shearlet transform, and the texture content of different image blocks is recognized at the same time of noise reduction. The edge-preserving noise reduction variational model is used to remove artificial artifacts from shearlet transform. In order to avoid the boundary effect brought by sub-graph block boundary, a new image continuation method based on the central affine transformation theory is proposed in this paper, which can effectively improve the image denoising effect. The experimental results show that the average peak signal to noise ratio (PSNR) of the proposed algorithm is 1 dB larger than that of the total variational model denoising algorithm. The average PSNR of the proposed algorithm is 2 dB larger than that of the curve wave denoising algorithm. Compared with the Shearlet algorithm based on Symmlet wavelet, the artifact in the field remote sensing image processed by this algorithm is reduced. When the standard deviation of Gao Si noise 蟽 is 10 ~ 20 and 30 dB, the peak signal-to-noise ratio (PSNR) is increased by 13.99% and 7.75%, respectively.
【作者单位】：中国农业大学信息与电气工程学院;北京邮电大学世纪学院移动媒体与文化计算北京市重点实验室;
【基金】：国家自然科学基金(41171337) “十二五”国家科技支撑计划项目(2015BAK04B01) 北京市自然科学基金(4172034)
【分类号】：TP751

【相似文献】