Image super-resolution method based on deep threshold convolutional neural network

Abstract

The invention discloses an image super-resolution method based on a deep threshold convolutional neural network, which is mainly used to solve the problem that the super-resolution effect of an image is reduced when the network is deepened in the prior art. The method comprises the following steps: (1) acquiring paired low-resolution and high-resolution image data as training data; (2), defining a threshold convolution layer to replace the existing convolution layer, and building an end-to-end deep threshold convolutional neural network; (3) inputting the training data to the deep threshold convolutional neural network to train the network through an Adam optimization method; and (4) using the trained deep threshold convolutional neural network to carry out image super-resolution. The gradient attenuation problem of the deep neural network is reduced. The application of the deep network in image super resolution is realized. The effect of image super-resolution is enhanced. The speed of image super-resolution is increased. The image super-resolution method can be used in fields such as satellite remote sensing, medicine, traffic monitoring and video compression.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to an image super-resolution method, which can be used in the fields of satellite remote sensing, medicine, traffic monitoring, video compression and the like. Background technique [0002] Image super-resolution refers to a technique that learns to restore a high-resolution image from a low-resolution image. Compared with low-resolution images, high-resolution images can express more detailed information, and its ability to express details is stronger. Therefore, image super-resolution has great applications in many fields, such as satellite remote sensing, medical fields, The field of traffic monitoring, and the field of video compression, etc. [0003] So far, there are three main categories of image super-resolution methods: interpolation-based, reconstruction-based, and learning-based methods. The learning-based method has made great progress in recent years and has ...

AI Technical Summary

Problems solved by technology

In the application process, since no additional learning parameters are needed, the time is short; but the disadvantage of this method is that the learning effect of the model will decrease when the number of layers of the network is deepened.

Images