Improved SRCNN for super-resolution reconstruction of retinal images
Yan Hui Lv, Hui Ma · 2021
Image super-resolution reconstruction means to recover high resolution image from low resolution image. It is widely used in satellite image, city monitoring, medical treatment and other fields. Medical image processing usually requires a high level of image detail. We designed an improved SRCNN for super-resolution reconstruction of retinal images (ISRCNN). In the image reconstruction part of this network, deconvolution was adopted for up-sampling to reconstruct high-resolution retinal images. Secondly, we deepened the layers of the network and adopted dense connections to improve the ability of network feature extraction. Finally, we adopted ReZero residue learning method, which could not only avoid the gradient disappearance or explosion caused by excessively deep network, but also help to reconstruct the detailed information of retinal image. We test models on DRIVE data sets, the results showed that our approach was effective in reconstructing retinal images.