去噪先验驱动的深度神经网络图像超分

Deep Neural Network for Image Super Resolution Driven by Prior Denoising

为了提高图像超分的质量,将双层卷积神经网络去噪模块嵌入到图像超分任务中,给出去噪先验驱动的深度神经网络图像超分方法。图像去噪模块可以快速灵活地处理不同水平的噪声,所提出的图像超分网络可以很好地实现图像超分复原。用训练好的网络对双三次下采样和高斯下采样的低分辨率图像进行超分测试,复原的高分辨率图像比使用其他图像超分方法具有较高的PSNR值和SSIM值。

In order to improve image super resolution,a double layer convolution neural network in image denoising is embedded in image restoration tasks.The image super resolution method driven by prior denoising with deep neural network is proposed.The image denoising model can deal with different noise levels quickly and flexibly.The proposed image super resolution network is able to achieve image restoration.The trained network tests the low resolution images generated by bicubic down-sampling.The restored super resolution images have higher PSNR values and SSIM values than other image super resolution image methods.