Endoscopic Image Deblurring and Super-Resolution Reconstruction Based on Deep Learning

There are two main reasons for the degradation of endoscopic image quality: 1) Motion blur; 2) Low imaging resolution. Since the blur kernels are highly nonlinear in real scenes, the restoration effect of the method of restoring motion blur by estimating the blur kernels is often not accurate enough. This paper proposes an end-to-end image blind deblurring algorithm based on convolutional neural network. This algorithm uses the architecture of combining image deblurring and super-resolution reconstruction of convolutional neural network, which divided into 3 parts: deblurring network, super-resolution network and feature fusion network. On the super-resolution task, this paper is based on densely connected convolutional networks (Dense-Net) [1], Res2Net [2] and segmentation channel method to improve network performance, and proposes different solutions for different types of image. Experimental results show that, compared with the previous method, the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of the restored image obtained by the algorithm are improved. The experimental results show that the restoration index and visual perception effect of the image obtained by this algorithm are improved compared with the previous method, and the algorithm greatly saves the computational cost during the training of the neural network.