Image recovery method and device based on pre-training auto-encoder

Abstract

The embodiment of the invention provides an image recovery method and device based on a pre-training auto-encoder. The method comprises the following steps: obtaining a speckle image; inputting the speckle image into an image restoration network model to obtain a restored image, wherein the image recovery network model comprises a first encoder sub-network and a first decoder sub-network which aretrained, the initial parameters of the first decoder sub-network are determined according to parameters in a pre-trained auto-encoder network model, and the auto-encoder network model is obtained bytraining according to a first training set, the image restoration network model is obtained by training according to a second training set, and the second training set comprises a second sample original image and a sample speckle image. According to the method, the auto-encoder network model is pre-trained, and the parameters in the trained auto-encoder network model are adopted to perform parameter initialization on the first decoder sub-network in the image recovery network model, so that training data set information can be fully utilized, and the convergence rate of the network during training is increased.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to an image restoration method and device based on a pre-trained self-encoder. Background technique [0002] In recent years, optical fibers have been widely used in imaging systems. Compared with traditional optical imaging systems using lenses, optical fibers have the advantages of small diameter, low loss, and low cost, and are an ideal intrusive light-guiding medium. Compared with ordinary optical fiber, multimode optical fiber allows simultaneous transmission of multiple modes, and has the characteristics of large core diameter and low coupling connection cost. [0003] However, since multimode fiber can transmit hundreds of guided wave modes, each mode has a different phase velocity, which will cause the original phase relationship between different image information to be lost, and the final output will output random speckle image. In fact, this process doe...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the embodiments of the present invention is to provide an image restoration method and device based on a pre-trained autoencoder to solve the problem of slow model convergence in existing methods for restoring original image information from speckle images transmitted by multimode optical fibers. Slow, technical issues with poor quality image recovery

Images