Application of convolutional neural networks in image super-resolution

Known for their strong learning abilities, convolutional neural networks (CNNs) have become mainstream methods for image super-resolution. However, substantial differences exist among deep learning methods of various types, and there is limited literature to summarize the relations and differences of different methods in image super-resolution. Thus, it is important to summarize such studies according to the loading capacity and the execution speed of devices. This paper first introduces the principles of CNNs in image super-resolution and then introduces CNN-based bicubic interpolation, nearest neighbor interpolation, bilinear interpolation, transposed convolution, subpixel layering, and meta-upsampling for image super-resolution to analyze the differences and relations of different CNN-based interpolations and modules. The performance of these methods is compared through experiments. Finally, this paper presents potential research points and drawbacks and summarizes the whole paper to promote the development of CNNs in image super-resolution.