Image restoration method based on lightweight feature pyramid model

Abstract

The invention discloses an image restoration method based on a lightweight feature pyramid model. The method comprises the following steps: constructing the lightweight feature pyramid model as an image restoration model; and inputting a to-be-restored image into the image restoration model, and outputting a restored image. According to the method, the image task is completed by using the lightweight pyramid repairing network, so that texture information and semantic information of the image can be well repaired while the model keeps low parameter quantity; the training process of the lightweight pyramid repair network model is faster, and the requirement for equipment required by training is lower; and the derivation process of the lightweight pyramid repair network model is faster, and the model can be deployed to a wider range of edge devices.

Description

technical field [0001] The invention belongs to the technical field of image restoration, in particular to an image restoration method based on a lightweight feature pyramid model. Background technique [0002] Image restoration refers to restoring the content of the missing part of the image through the known information in the image. In terms of academics, academic research in foreground removal, background modeling, video analysis, multi-view synthesis, face reconstruction, etc. will involve image restoration, which has fundamental research value in graphics and imaging. In terms of application, it has a wide range of application values ​​in the fields of image editing, film and television special effects production, virtual reality and digital cultural heritage protection. The research motivation of image restoration originated from the manual restoration of incomplete artistic images. Therefore, the early image restoration methods are very similar to the artificial res...

AI Technical Summary

Problems solved by technology

This kind of generator will not be able to restore the detailed texture information and macro semantic information of the image well

In the research field of other visual tasks, although the network structure of multi-scale representation such as feature pyramid network can have better performance, the network cannot be directly applied to image restoration tasks, and has a huge amount of parameters.

At present, the application of deep learning in image restoration is still in its infancy, and there are still many problems, such as: the semantic information and texture details of the repaired image cannot match the real image; the training time of the deep learning model is too long; Deriving predictions takes too long, the number of model parameters is too large, etc.

Images