为有效抑制椒盐噪声对图像信息的影响,根据椒盐噪声随机破坏图像中像素值的显著特征,本文提出一种耦合噪声检测的自适应模糊正则化噪声去除模型。一方面,基于L_(1)范数建立数据保真项,实现对图像统计分布进行有效拟合。另一方面,通过对...为有效抑制椒盐噪声对图像信息的影响,根据椒盐噪声随机破坏图像中像素值的显著特征,本文提出一种耦合噪声检测的自适应模糊正则化噪声去除模型。一方面,基于L_(1)范数建立数据保真项,实现对图像统计分布进行有效拟合。另一方面,通过对图像中像素相似性的有效量化实现图像中噪声的检测,并将此耦合至正则项中,使得模型可依据像素点实际受噪声的污染对其施加惩罚程度,最终实现椒盐噪声的自适应模糊去除。本文采用交替方向乘子法(Alternating direction method of multipliers,ADMM)进行模型的数值结果实现,并运用峰值信噪比(Peak signal-to-noise ratio,PSNR)及结构相似性(Structural similarity,SSIM)对实验结果进行评定。实验结果表明,本文提出的模型在PSNR及SSIM方面得到显著提升,其中对于灰度图像的去噪实验PSNR最高可提高1.3dB,SSIM最高可提高0.2。展开更多
In this paper, the interferences of X-ray image noise on a bone age model, Xception model, were studied. We conduct a comparative experiment test according to the output performance of the neural network model using b...In this paper, the interferences of X-ray image noise on a bone age model, Xception model, were studied. We conduct a comparative experiment test according to the output performance of the neural network model using both the original image training and noise-added (Gaussian noise plus salt-pepper noise) training, and analyze the anti-interference ability of the Xception model, hoping to improve it through noise enhancement training and generalize the application ability of the model. The results show that the model trained with noise-added (Gaussian noise plussalt-pepper noise) images can make predictions that are more robust and less affected by the image disturbances, such as image noise.展开更多
文摘为有效抑制椒盐噪声对图像信息的影响,根据椒盐噪声随机破坏图像中像素值的显著特征,本文提出一种耦合噪声检测的自适应模糊正则化噪声去除模型。一方面,基于L_(1)范数建立数据保真项,实现对图像统计分布进行有效拟合。另一方面,通过对图像中像素相似性的有效量化实现图像中噪声的检测,并将此耦合至正则项中,使得模型可依据像素点实际受噪声的污染对其施加惩罚程度,最终实现椒盐噪声的自适应模糊去除。本文采用交替方向乘子法(Alternating direction method of multipliers,ADMM)进行模型的数值结果实现,并运用峰值信噪比(Peak signal-to-noise ratio,PSNR)及结构相似性(Structural similarity,SSIM)对实验结果进行评定。实验结果表明,本文提出的模型在PSNR及SSIM方面得到显著提升,其中对于灰度图像的去噪实验PSNR最高可提高1.3dB,SSIM最高可提高0.2。
文摘In this paper, the interferences of X-ray image noise on a bone age model, Xception model, were studied. We conduct a comparative experiment test according to the output performance of the neural network model using both the original image training and noise-added (Gaussian noise plus salt-pepper noise) training, and analyze the anti-interference ability of the Xception model, hoping to improve it through noise enhancement training and generalize the application ability of the model. The results show that the model trained with noise-added (Gaussian noise plussalt-pepper noise) images can make predictions that are more robust and less affected by the image disturbances, such as image noise.
