Image Super-resolution Reconstruction Algorithm Based on Convolutional Neural Network
Jingxuan He, Jian Zhang, Zhang Yonghui, Wang Rong · 2018
Aiming at the problem that the existing SRCNN algorithm has too long training time, poor reconstruction performance and slow running speed, a new image super-resolution reconstruction algorithm based on convolutional neural network is proposed. The algorithm uses low-resolution images as the network input, the higher-order representation of the image is learned using the convolution operation, the high-resolution image is up-sampling by the deconvolution operation, and the residual structure is added to the network, so that the entire network can converge better. Experimental results on Set5, Set14, and BSD200 test sets show that compared with Bicubic, SRCNN and other methods, the method proposed in this paper is more efficient on super-resolution reconstruction of images and the convergence speed of the network is faster.