Image Super-Resolution Reconstruction Based on Multi-scale Convolutional Neural Network

For image super-resolution based on convolutional neural network, there are many problems such as large amount of calculation, many parameters, and unresolved images. This paper proposes an image super-resolution reconstruction algorithm based on multi-scale convolutional neural network. The multi-scale convolution kernel method is introduced into convolutional neural networks. Multi-scale feature extraction is achieved for different sizes of convolutional layers, at the same time, the learning parameters are improved and the network parameters are reduced. Maxout is used as an activation function to introduce competing elements. At the same time, the Skip Connection in the residual network is added to the network model to accelerate the training of deep neural networks. Experiments show that the subjective visual and objective evaluation of this algorithm has been improved to a certain extent. The edge effect of the reconstructed high-resolution image is more clear when reducing the number of network parameters, and more detailed image information is recovered.