Microscope Image Deblurring Method

With the development of computer technology,pathologists have realized the low efficiency of traditional microscopic pathological diagnosis methods,which has led to the large-scale development of digital pathological diagnosis.Obtaining clear images through rapid scanning technology is crucial in digital pathology scanning.There are two main ways to obtain clear images by image processing methods: one is to use image fusion technology to fuse multi-focus images into clear images.The other is image deblurring technology,the deblur image is restored to a clear image.In common multi-focus image fusion methods,the whole image is added into the fusion operation,which has high time complexity.In image deblurring,non-blind deblurring is carried out when the fuzzy kernel is known,but in the actual application scene,the fuzzy kernel is often unknown,so the problem of blind deblurring is more researchable.The above blind deblurring method has a good application in the field of natural image and moving image,but it still has some shortcomings in the field of microscope image.In order to solve the problems of image fusion and image deblurring,this paper first proposes an image sharpness evaluation method based on convolutional neural network,which can effectively and quickly evaluate whether the image is clear.Then,a method of TCT(Thinprep Cytologic Test)cell deblurring based on image fusion is proposed,and the image content is analyzed to achieve the purpose of fast image fusion.Finally,a defocus image deblurring method based on Deblur GAN is proposed for regional blurred images and defocused images.The work of this paper includes the following aspects:1.An image sharpness evaluation method based on convolutional neural network is proposed.This method is to train a regression model of sharpness estimation by using a large amount of data.According to the output score of the model,the clarity of image can be determined.In addition,the sharpness label function is proposed,which is superior to other methods in unimodality,precisi on and sharpness.The function value is input into the regression model as the sharpness label in the training,and get the sharpness prediction results of the unreferenced image.Finally,the sharpness evaluation method is applied to the focusing field of microscope to reduce the focusing times and improve the focusing efficiency.2.A method of TCT cell deblurring based on image fusion is proposed.In this method,multiple images are obtained by quick image acquisition,and then multi-focus images are screened out.Then the image is segmented,and the segmentation results are divided into three categories: single cell,cluster cell and impurity by VGG-16(Visual Geometry Group-16)classifier model.For single cells,images were selected by contrast sharpness method for fusion;Cluster cells were fused by guided filtering;Do not dispose of impurity.Finally,the fused cell content is stitched into a clear image as the final result.3.A method of defocus image deblurring based on Deblur GAN is proposed.In this method,attention module is added into the Res Net residual block.Attention mechanism was introduced to strengthen the repair effect of cell edge and weaken the influence of background on the deblurring effect.Under the premise of ensuring efficiency,enhance the effect of image deblurring.Experimental results show that the method prese nted in this paper can obtain clear microscope images on the basis of judging image sharpness.I n addition,it can reduce the focusing time of the microscope,improve the efficien cy of image fusion,and solve the problem of image deblurring and edge repair,so can get clear microscope image.