Research And Application Of Speckle Denoising Method Based On FFDNet

Speckle noise is widely found in images in the fields of optical detection and medical imaging,such as electronic speckle pattern interferometry(ESPI)fringe patterns and optical coherence tomography(OCT)retinal images.The presence of speckle noise reduces the quality of the image and severely affects the acquisition of useful information.Therefore,it is very meaningful to study the removal of speckle noise.In actual engineering,it is often necessary to denoise multiple images.The traditional method needs to be processed frame by frame,which is inconvenient and time consuming.The rise of deep learning technology has provided new research ideas for different research fields.Fast and Flexible Denoising Network(FFDNet)is an advanced method for denoising natural images based on deep learning.In this paper,we will conduct speckle denoising based on FFDNet and extend it to the speckle noise removal of ESPI fringe patterns and OCT retinal images.The details are as follows:(1)In this paper,we conduct research on denoising methods based on FFDNet.The network introduces the idea of residual learning,constructs a residual map to learn the noise distribution in the original image through a multi-layer convolutional neural network.Denoising results can be obtained by subtracting the noise learned from the network using the original image.(2)In view of the need for large-scale fringe pattern processing in dynamic electronic speckle pattern interferometry measurement,this paper proposes a FDDNetbased ESPI fringe pattern batch denoising method.A new computer simulation method was used to create a training data set for the ESPI fringe pattern denoising network.In this paper,a set of simulated fringe patterns and two sets of experimental ESPI fringe patterns are tested.The experimental results show that the proposed method does not require complex parameter adjustment in the process of removing speckle noise,and is especially suitable for processing a large number of multi-frame ESPI.Striped image.In this paper,the proposed ESPI fringe pattern denoising method is applied to ESPI dynamic thermal deformation measurement,and the batch denoising of ESPI fringe pattern in dynamic measurement is realized and the corresponding phase is extracted.The effectiveness of the method is verified.(3)In this paper,we propose an OCT retinal image denoising method based on FFDNet.We compared the denoising results with three representative methods.We tested the proposed method using a total of 100 OCT retinal images from 4 different pathological conditions and a set of OCT retinal images acquired by an ocular patient during treatment.Experiments show that the OCT retinal image denoising method proposed in this paper can effectively reduce the speckle noise while protecting the structural information such as image edges and textures,and does not require complex parameter adjustment.In addition,we applied the OCT retinal denoising method to the OCT retinal image classification.By comparing the accuracy of the classification before and after denoising,the effectiveness and practicability of the proposed OCT retinal image denoising method were further verified.