Super-Resolution Based on Noise Resistance Deep Convolutional Network
Hengjian Li, Yunxing Gao, Jiwen Dong, Guang Feng · 2018
In this paper, we present a novel deep network model which is designed to deal with medical image super-resolution and has some resistance to noise contamination. We are mainly aimed at the medical image susceptible to noise contamination in the collection and transmission process, and the noise in medical images will be amplified after super-resolution reconstruction. We improve the Super-Resolution Convolution Neural Network (SRCNN) model mainly in two aspects. First, in order to make our model with noise resistance, we use discrete Harr wavelet transform as preprocessing algorithm. Second, we use adaptive partition algorithm based on image content to block the original image which can reduce the time complexity. The experimental results show that our model still achieves a good objective evaluation index (PSNR) and subjective visual effect on medical images that add Gaussian white noise. Our model is fast and effective and also have important guiding significance for the difficulty and risk assessment of surgical feasibility.