Research On Automatic Focusing And Mosaic Algorithms Of Pathological Microscopic Images

Microscope as commonly used equipment in medical field,but its depth of field and vision are limited.Pathologists need to constantly adjust the lens to achieve focus when observing pathological images.At the same time,they also need to move the slide to observe the various positions of pathological images to complete the diagnosis.This traditional way will lead to a large workload,inefficiency and low diagnostic accuracy of doctors,and it is difficult to meet the current needs.At this time,computer science and digital image processing technology have made great progress.In order to collect clear images,automatic focusing technology can be applied.Two or more small field of view images with overlapping parts can be assembled into a large field of view image by image mosaic technology.In this way,not only the field of vision is wider,but also the image clarity is higher,which can facilitate the observation and diagnosis of pathologists and reduce the burden of doctors.In this paper,the auto-focusing algorithm and its mosaic algorithm for microscopic images are analyzed and studied.The specific contents of the research are as follows:1.The basic working method and principle of the microscope are introduced.On this basis,the principle and method of autofocus correlation is described.The research of image processing method is the key point.For the focusing method of image processing,we understand how to use edge features to complete image focusing,and find out the shortcomings of this algorithm in the auto-focusing process of microscopic image.Therefore,we propose an auto-focusing method of microscopic image combining with directional gradient histogram(HOG)features.Combining HOG features with Laplace gradient to judge whether the image is clear or not,and then auto-focusing of microscopic image is realized.The experimental results show that when the object is small and the background isrelatively complex,the improved algorithm can still find the clearest image accurately.2.For image matching based on feature points,Harris feature points extraction,SUSAN feature points extraction and SIFT feature points extraction are introduced.Firstly,the principle of SIFT feature is introduced in detail,and then a SIFT image matching method combining two-dimensional entropy is proposed.In the feature extraction stage,some unstable feature points are eliminated by calculating the two-dimensional entropy of the image.In the feature point matching stage,by judging the difference between the reference image and the feature points in the image to be matched,if the two-dimensional entropy difference of the feature point pair is too large,the subsequent calculation of the Euclidean distance ratio for the matching pair will not be carried out,which greatly saves the calculation time.Experiments show that the matching speed of the traditional SIFT algorithm is slow.The improved algorithm improves the matching speed greatly on the basis of maintaining the robustness of matching.3.After finding the matching pair between the reference image and the image to be matched,the stitching result of the image is obtained by using H-space transformation matrix.However,the mosaic images always have some mosaic marks,so it is necessary to use image fusion technology to achieve seamless mosaic of images.The direct average fusion,weighted average fusion and gradual-in and gradual-out fusion algorithms are introduced respectively,and an improved fusion algorithm based on partial differential equation is proposed.The experimental results show that the improved fusion algorithm is more effective in removing image mosaic marks and suppressing noise.