在地震图像采集过程中,由于存在各种影响因素,采集到的地震资料常常夹杂着各种噪声,严重影响了后续研究的准确性。因此,图像去噪成为至关重要的一步。自适应中值滤波作为一种经典的去噪方法,在去除椒盐噪声方面表现优异。近来,一种结合...在地震图像采集过程中,由于存在各种影响因素,采集到的地震资料常常夹杂着各种噪声,严重影响了后续研究的准确性。因此,图像去噪成为至关重要的一步。自适应中值滤波作为一种经典的去噪方法,在去除椒盐噪声方面表现优异。近来,一种结合传统全变差与自监督学习网络模型的方法——S2S-WTV被提出,该方法不仅拥有神经网络的强大表示能力,还有传统模型的泛化能力,在针对高斯噪声去除方面展现出了出色的性能。基于此,本文提出了一种结合传统方法与神经网络的多种噪声去噪模型,显著提升了去噪后的视觉效果与图像结构相似性。During seismic image acquisition, various factors often result in the data being noisy, severely impacting the accuracy of subsequent studies. Therefore, image denoising is critical. Adaptive median filtering, a classic denoising technique, excels at removing salt and pepper noise. Recently, S2S-WTV, a method that integrates traditional total variation with self-supervised learning network models, has been introduced. This approach combines the strong representational capabilities of neural networks with the generalization abilities of traditional models, showing outstanding performance in reducing Gaussian noise. Building on this, the paper introduces a multi-noise denoising model that combines traditional methods with neural networks, significantly enhancing visual quality and image structure similarity post-denoising.展开更多
针对自适应中值滤波在窗口迭代过程中存在像素点重复参与运算导致算法复杂度较高的问题,提出了一种改进的中值滤波算法.首先依据有效像素点与窗口中心点间的坐标距离来快速确定最佳滤波窗口尺寸,避免了窗口迭代造成的像素点重复排序;之...针对自适应中值滤波在窗口迭代过程中存在像素点重复参与运算导致算法复杂度较高的问题,提出了一种改进的中值滤波算法.首先依据有效像素点与窗口中心点间的坐标距离来快速确定最佳滤波窗口尺寸,避免了窗口迭代造成的像素点重复排序;之后对窗口内的有效像素点进行取中值操作,有效削弱了噪声点的干扰,进一步提升了图像滤波的质量.经实验验证,与自适应中值滤波算法比较,复杂度显著降低,峰值信噪比(peak signal to noise ratio,PSNR)值平均提高10 d B左右.和同类文献比较在算法复杂度和图像降噪效果间做出了一个较佳的权衡.最后将该算法应用于Kinect深度图降噪上获得了不错的效果.展开更多
文摘在地震图像采集过程中,由于存在各种影响因素,采集到的地震资料常常夹杂着各种噪声,严重影响了后续研究的准确性。因此,图像去噪成为至关重要的一步。自适应中值滤波作为一种经典的去噪方法,在去除椒盐噪声方面表现优异。近来,一种结合传统全变差与自监督学习网络模型的方法——S2S-WTV被提出,该方法不仅拥有神经网络的强大表示能力,还有传统模型的泛化能力,在针对高斯噪声去除方面展现出了出色的性能。基于此,本文提出了一种结合传统方法与神经网络的多种噪声去噪模型,显著提升了去噪后的视觉效果与图像结构相似性。During seismic image acquisition, various factors often result in the data being noisy, severely impacting the accuracy of subsequent studies. Therefore, image denoising is critical. Adaptive median filtering, a classic denoising technique, excels at removing salt and pepper noise. Recently, S2S-WTV, a method that integrates traditional total variation with self-supervised learning network models, has been introduced. This approach combines the strong representational capabilities of neural networks with the generalization abilities of traditional models, showing outstanding performance in reducing Gaussian noise. Building on this, the paper introduces a multi-noise denoising model that combines traditional methods with neural networks, significantly enhancing visual quality and image structure similarity post-denoising.
文摘针对自适应中值滤波在窗口迭代过程中存在像素点重复参与运算导致算法复杂度较高的问题,提出了一种改进的中值滤波算法.首先依据有效像素点与窗口中心点间的坐标距离来快速确定最佳滤波窗口尺寸,避免了窗口迭代造成的像素点重复排序;之后对窗口内的有效像素点进行取中值操作,有效削弱了噪声点的干扰,进一步提升了图像滤波的质量.经实验验证,与自适应中值滤波算法比较,复杂度显著降低,峰值信噪比(peak signal to noise ratio,PSNR)值平均提高10 d B左右.和同类文献比较在算法复杂度和图像降噪效果间做出了一个较佳的权衡.最后将该算法应用于Kinect深度图降噪上获得了不错的效果.
