A picture texture enhancement super-resolution method based on a deep feature translation network

Abstract

The invention relates to a picture texture enhancement super-resolution method based on a deep feature translation network, and belongs to the technical field of computer vision. The method comprisesthe steps of firstly, processing the training data, and then designing a network structure model which comprises a super-resolution reconstruction network, a fine-grained texture feature extraction network and a discrimination network; and then, designing a loss function for training the network by adopting a method of combining various loss functions, and training the network structure model by using the processed training data to obtain a super-resolution reconstruction network with a texture enhancement function, and finally, inputting the low-resolution image into the super-resolution reconstruction network, and performing reconstruction to obtain a high-resolution image. According to the method, the picture texture information can be extracted under finer granularity, a mode of combining multiple loss functions is adopted, compared with other methods, the method guarantees that the original picture is loyalty, the texture feature information can be recovered, and the picture is clearer. The method is suitable for any picture, has a good effect, and has good universality.

Description

technical field [0001] The present invention relates to a picture texture enhancement super-resolution method based on a deep feature translation network, in particular to a super-resolution method based on a convolutional neural network called a deep feature translation network and using various loss function training to enhance picture texture information. A resolution method belongs to the technical field of computer vision. Background technique [0002] In the Internet age, there are a lot of low-resolution pictures. In addition, many high-resolution images are compressed during transmission because of their large size and storage space, resulting in low-resolution images. Low-res images have poor quality and are too small in size. Using the super-resolution method, low-resolution images can be reconstructed into high-resolution images, so it has broad application prospects in military, medical, education and many other fields. [0003] Traditional super-resolution me...

AI Technical Summary

Problems solved by technology

However, the pictures obtained based on the method of generating confrontation network alone often have the shortcomings of more noise, single texture and infidelity to the original picture.

Images