A watershed-based cervical liquid-based cell segmentation method and system

Abstract

The application discloses a kind of based on watershed cervical liquid-based cell segmentation method and system, it is related to cell segmentation technical field.The steps of including image pre-processing, preliminary rough segmentation of watershed, Kmeans target classification and the target optimization of over-segmentation and under-segmentation, solve the problem that threshold segmentation method is relatively inflexible and the segmentation ability of limited and deep learning segmentation method needs a lot of annotation data.For under-segmentation problem of watershed method, the method of secondary thresholding solves the problem of concave point detection and ellipse fitting;For over-segmentation problem, the method of inflation is used to judge again to reduce the rate of missed detection.The cervical liquid-based cell segmentation method of the application can obtain good segmentation effect for cells in cervical cell image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cell segmentation, and in particular to a cervical liquid-based cell segmentation method and system based on watershed. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] When a pathologist performs pathological diagnosis on a cervical liquid-based cell digital slice, not only the pathological area in the cervical liquid-based cell slice needs to be found, but also the number of granulocytes such as neutrophils, lymphocytes and phagocytes needs to be combined to comprehensively judge the cell inflammation. Therefore, the accuracy of cell segmentation of the digital slice has a great influence on the result of pathological diagnosis.

[0004] When cells in a digital image are segmented and counted, commonly used methods include threshold methods, including a global threshold method, which often sets a global threshold to segment the entire image according to the gray value, and has no constraint condition for the integrity of a single target; an OTSU threshold method, which obtains a threshold according to a gray value histogram of the image, and is more suitable for images with a bimodal gray value, and is not ideal for multi-layer color transformation; and an adaptive threshold method, which selects a threshold for a region, and is often good for target edge and contour extraction, but is poor for global images. At present, the watershed method is more commonly used, which regards the image as topological landforms in geodesy, and slowly expands the influence domain of each local minimum value to segment the image. However, the inventors find that the existing watershed method can only complete a rough segmentation, often resulting in under-segmentation and over-segmentation, and has insufficient segmentation capacity. In addition, the image segmentation method based on deep learning seriously depends on a large amount of labeled data, and is difficult to label in the case of dense distribution and large number of granulocytes, and has poor applicability. SUMMARY

[0005] In view of the deficiencies in the prior art, the present application aims to provide a cervical liquid-based cell segmentation method and system based on watershed, which first pre-processes and enhances the image, then uses the watershed method to complete preliminary rough segmentation of the granulocytes in the image to obtain the general target of the granulocytes, and finally optimizes the fine segmentation of the under-segmentation and over-segmentation problems in the rough segmentation image to correct the segmentation effect of the image.

[0006] In order to achieve the above-mentioned purpose, the present application is realized by the following technical scheme:

[0007] The present application provides a cervical liquid-based cell segmentation method based on watershed in the first aspect, comprising the following steps:

[0008] Step S1: Obtain a cervical liquid-based cell image, and perform grayscale, thresholding, and opening operation preprocessing operations on the cervical liquid-based cell image;

[0009] Step S2: Calculate foreground markers, background markers, and unknown regions in the image, and use a watershed method to complete preliminary coarse segmentation;

[0010] Step S3: Obtain all target contours of the coarse segmentation, and calculate related information of each target, and use a Kmeans method to divide the cells into under-segmentation, over-segmentation, and final targets;

[0011] Step S4: Optimize the over-segmentation targets and under-segmentation targets, and add the optimized targets to the final targets.

[0012] Further, the preprocessing operation specific steps are: performing grayscale processing on the cervical cell original image to obtain a grayscale image: the RGB color mode used by the original image is represented by three color variables of red, green, and blue, and each pixel of the RGB component is assigned an intensity value in the range of 0-255, which is converted into a single-channel image with only one gray value intensity component and a gray range of 0-255 through grayscale conversion;

[0013] Performing binarization operation on the grayscale image to obtain a binary image: converting the grayscale image with 0-255 pixel values into a black and white image with only 0 and 255 values;

[0014] Performing denoising processing on the binary image: using morphological opening operation, i.e., first eroding and then expanding the binary image.

[0015] Further, the specific steps of calculating the foreground markers, background markers, and unknown regions in the image include:

[0016] Obtaining the background markers of the image, performing morphological dilation processing on the target of interest in the binary image, expanding the boundary points of the binarization object, merging all background points in contact with the object into the object, and making the boundary of the target of interest in the image expand outward, so that the background markers contain more true target of interest;

[0017] Obtaining the foreground markers of the image, performing distance transformation operation on the binary image, calculating the distance of each non-zero point in the image from the nearest zero point, converting the gray value matrix of the binary image into a distance matrix to the nearest zero point, and obtaining the foreground markers of the image by binarizing the distance matrix;

[0018] Obtaining the unknown region markers of the image, subtracting the foreground marker matrix from the background marker matrix to obtain the unknown region marker matrix of the image.

[0019] Further, the specific steps of using the watershed method to complete the preliminary rough segmentation include: marking the unknown region in the foreground label as a 0 pixel value through connected domain analysis;

[0020] The original image is segmented using the watershed method to segment the target of interest, the contour of the foreground label is transmitted as a seed, other pixel points on the image are judged according to the rules of the watershed algorithm, and the region belonging of each pixel point is demarcated to obtain all the targets of interest of the cells, and the initial rough segmentation is completed.

[0021] Further, the related information of each target includes: area, perimeter, convex hull, circumscribed circle area, circular granularity, solidity, major axis, minor axis, solidity, ellipticity, width, height and eccentricity.

[0022] Further, the specific steps of using the Kmeans method to divide the cells into under-segmentation, over-segmentation and final targets are as follows:

[0023] Randomly select 3 samples as initial cluster centers;

[0024] For each sample in the data set, calculate its distance to the k cluster centers, and attribute it to the class corresponding to the cluster center with the smallest distance;

[0025] For each cluster, recalculate its cluster center position;

[0026] Repeat the iteration operation until the maximum number of iterations is reached or the cluster center position is unchanged.

[0027] Further, the over-segmented target is optimized using secondary thresholding, and the under-segmented target is optimized using dilation processing.

[0028] Further, the under-segmented target is optimized, specifically including:

[0029] Crop the under-segmented target region, calculate the circumscribed rectangle coordinates of each under-segmented target, and crop the under-segmented target image from the original cervical cell image;

[0030] Eliminate the background of the under-segmented region image, fill the non-over-segmented part in the cropped over-segmented rectangular image with (0, 0, 0) pixel value, only keep the over-segmented target, and eliminate other targets in the image;

[0031] Grayscale processing of the under-segmented region image, converting the RGB three-channel image to a single-channel grayscale image;

[0032] Secondary thresholding of the under-segmented region image, converting the grayscale image to a binary image;

[0033] The binary image is subjected to a target searching operation, and whether the area of each target is in the interval [40, 1500] is judged, if yes, the final target is added.

[0034] Further, the over-segmented targets are optimized, and the optimization specifically includes:

[0035] The over-segmented target region is cropped, the circumscribed rectangle coordinates of each over-segmented target are calculated, and the surrounding region of the target is widened, and the over-segmented target image is cropped from the original cervical cell image;

[0036] The over-segmented region image is subjected to a gray-scale processing, and the RGB three-channel image is converted into a single-channel gray-scale image;

[0037] The gray-scale image is subjected to a threshold processing, and the gray-scale image is converted into a binary image;

[0038] The binary image is subjected to an inflation processing, and the under-segmented target boundary in the binary image is expanded outward; if two objects are close to each other, the two objects will be connected together.

[0039] The binary image is subjected to a target searching operation, and whether the area of each target is in the interval [40, 1500] is judged, if yes, the final target is added.

[0040] The second aspect of the present application provides a cervical liquid-based cell segmentation system based on a watershed, comprising:

[0041] A preprocessing module is configured to acquire a cervical liquid-based cell image, and perform a gray-scale processing, a threshold processing and an open operation preprocessing operation on the cervical liquid-based cell image;

[0042] A first segmentation module is configured to calculate foreground markers, background markers and unknown regions in the image, and complete a preliminary rough segmentation by using a watershed method;

[0043] A second segmentation module is configured to obtain all target contours of the rough segmentation, calculate the area, the perimeter, the circularity and the solidity information of each target, and divide the cells into under-segmented targets, over-segmented targets and final targets by using a Kmeans method;

[0044] A target optimization module is configured to optimize the over-segmented targets and the under-segmented targets, and add the optimized targets to the final targets.

[0045] The beneficial effects of the embodiments of the present application are as follows:

[0046] The application uses a watershed algorithm to preliminarily coarsely segment the cervical cell image, solves the problems of limited segmentation ability of a relatively rigid threshold segmentation method and the need of a large amount of labeled data for a deep learning segmentation method; for the under-segmentation problem of the watershed method, the method of secondary thresholding solves the problems of concave point detection and ellipse fitting; for the over-segmentation problem, the method of inflation is used to judge again to reduce the false detection rate. The cervical liquid-based cell segmentation method can obtain good segmentation effect on the cells in the cervical cell image, and can play a good auxiliary role in the judgment of cervical liquid-based cell inflammation in the clinic. BRIEF DESCRIPTION OF DRAWINGS

[0047] The drawings accompanying the specification of this application form a part thereof, serve to provide further understanding of the application, and together with the description of the exemplary embodiments of the application and the explanation thereof, make an explanation of the application, and do not constitute an improper limitation of the application.

[0048] Figure 1 A flowchart of a cervical cell segmentation method of the embodiment one of the application;

[0049] Figure 2 A framework structure diagram of a cervical cell segmentation process of the embodiment one of the application. DETAILED DESCRIPTION

[0050] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0051] It should be noted that the terms used herein are only for the purpose of describing the specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that when the terms "comprise" and / or "include" are used in the specification, there is a presence of the features, steps, operations, devices, components and / or their combinations;

[0052] Embodiment one:

[0053] The embodiment one of the application provides a cervical liquid-based cell segmentation method based on a watershed, as shown in the figure, including image preprocessing, watershed preliminary coarse segmentation, Kmeans target classification, and target optimization of over-segmentation and under-segmentation, and the specific process is: Figure 1

[0054] Step S1: Obtain the cervical liquid-based cell image, and perform gray scale, thresholding, and open operation preprocessing operations on the cervical liquid-based cell image.

[0055] ​Preferably, the cervical liquid-based cell image is a digital slice image scanned by a pathological slice scanner on a cervical liquid-based slide.

[0056] S1.1: Grayscale processing of the cervical cell original image to obtain a grayscale image: the RGB color mode used by the original image is represented by three color variables of red, green and blue (R, G, B) for each pixel point, and the RGB components of each pixel are assigned an intensity value in the range of 0-255. Through grayscale conversion, a single-channel image with only one gray value intensity component and a gray range of 0-255 is obtained. This makes the image contrast enhanced and the features more obvious, thereby improving the display effect of the image and making it easier to process.

[0057] S1.2: Binaryzation of the grayscale image to obtain a binary image: the grayscale image with pixel values of 0-255 is converted into a black-and-white image with only two values of 0 and 255, which is conducive to highlighting the outline of the target of interest and more conducive to subsequent segmentation and extraction of the target of interest.

[0058] S1.3: Denoising of the binary image: this embodiment uses morphological opening operation, i.e. erosion followed by dilation, to process the binary image. The operation can effectively remove isolated points and small area objects in the image to achieve the effect of denoising; and can eliminate the details in the image that are smaller than the structural elements, thereby smoothing the object outline and keeping the local shape unchanged.

[0059] Step S2: calculating the foreground marker, background marker and unknown region in the image, and using the watershed method to complete the preliminary rough segmentation. S2.1: obtaining the background marker of the image, and performing morphological dilation processing on the target of interest in the binary image to expand the boundary points of the binary object, merging all background points in contact with the object into the object, so that the boundary of the target of interest in the image expands outward, and the background marker contains more of the true target of interest.

[0061] S2.2: obtaining the foreground marker of the image, performing distance transformation operation on the binary image obtained in step S1 to calculate the distance of each non-zero point in the image from the nearest zero point, converting the gray value matrix of the binary image into a distance matrix to the nearest zero point, and the matrix reflects the distance relationship between each pixel and the background so as to extract the foreground. By binaryzation of the distance matrix, the foreground marker of the image is obtained. In particular, the threshold value of binaryzation is obtained from the maximum value in the distance matrix:

[0062] Threshold dis =α×max(distance)

[0063] wherein, Threshold disFor the binary threshold, a is the threshold coefficient, in this example a = 0.1, and distance is the distance matrix.

[0064] S2.3: Obtain the image unknown region label, subtract the foreground label matrix from the background label matrix to obtain the image unknown region label matrix.

[0065] Mark the unknown region in the foreground label as 0 pixel value by connected component analysis.

[0066] S2.4: Use the watershed method to segment the target of interest in the original image, use the contour of the foreground label as the seed, and judge other pixel points on the image according to the rules of the watershed algorithm, and determine the region attribution of each pixel point to obtain all the targets of interest of the cells, and complete the initial rough segmentation.

[0067] Preferably, the watershed segmentation algorithm is a mathematical morphological segmentation method based on topological theory, and the basic idea is to regard the intensity value of the cervical cell original image as a topological landform in geodesy, and slowly expand the influence domain of each local minimum value (i.e. the foreground label) outward, build a dam at the confluence of two catchment basins, i.e. form a watershed to complete the image segmentation.

[0068] Step S3: Obtain the contour of all the targets of rough segmentation, and calculate the related information of each target, and use the Kmeans method to divide the cells into under-segmentation, over-segmentation and final target.

[0069] S3.1: For the target of interest segmented by the watershed method, calculate the area, perimeter, convex hull, circumscribed circle area, circular granularity, solidity, major axis, minor axis, solidity, ellipticity, width, height, and eccentricity of each target, encode the information into a one-hot form feature vector, define three cluster categories of under-segmentation, over-segmentation and final target, and use the Kmeans method to divide the cells into under-segmentation, over-segmentation and final target.

[0070] S3.2: The specific steps of Kmeans are as follows:

[0071] Randomly select 3 samples as initial cluster centers;

[0072] Calculate the distance of each sample in the data set to the k cluster centers, and attribute it to the class corresponding to the cluster center with the smallest distance;

[0073] Recalculate the cluster center position of each cluster;

[0074] Repeat the above two steps until the number of iterations reaches the maximum value or the cluster center position is unchanged.

[0075] Step S4: Optimize the over-segmented targets and under-segmented targets, and add the optimized targets to the final targets.

[0076] S4.1: Optimize the over-segmented targets using secondary thresholding.

[0077] S4.1.1: Crop the under-segmented target area, calculate the coordinates of the bounding rectangle of each under-segmented target, and crop the under-segmented target image from the original cervical cell image.

[0078] S4.1.2: Eliminate the background of the under-segmented region image, fill the non-over-segmented part of the cropped over-segmented rectangular image with (0, 0, 0) pixel value, and only keep the over-segmented target, eliminating other targets that are cropped together in the image.

[0079] S4.1.3: Grayscale processing of under-segmented region image, convert RGB three-channel image to single-channel grayscale image.

[0080] S4.1.4: Secondary thresholding of under-segmented region image, convert grayscale image to binary image. Unlike the binary operation in step S1, the threshold of this binary operation is reduced by one gradient, making the under-segmented targets more prominent than in step S1. In particular:

[0081] Threshold2 = Threshold1 - Gradient

[0082] Threshold2 is the second thresholding threshold, Threshold1 is the first thresholding threshold, and Gradient is the reduced gradient. In this embodiment, Gradient = 60.

[0083] S4.1.5: Perform target search operation on the binary image, determine whether the area of each target is in the interval [40, 1500], if yes, add it to the final target; if not, discard it.

[0084] S4.2: Optimize the under-segmented targets using dilation processing.

[0085] S4.2.1: Crop the over-segmented target area, calculate the coordinates of the bounding rectangle of each over-segmented target, and widen the target area, crop the over-segmented target image from the original cervical cell image.

[0086] S4.2.2: Grayscale processing of over-segmented region image, convert RGB three-channel image to single-channel grayscale image.

[0087] S4.2.3: Thresholding of grayscale image, convert grayscale image to binary image.

[0088] S4.2.4: Perform dilation operation on the binary image to expand the under-segmented object boundary in the binary image outward. If two objects are too close to each other, the two objects will be connected together. In this embodiment, for the over-segmentation phenomenon that a cell in the image is segmented into two cells, the dilation operation will expand the cell boundary to recombine the two cells together.

[0089] S4.2.5: Perform a search target operation on the binary image to determine whether the area of each target is in the interval [40, 1500]. If yes, the target is added to the final target. If no, the target is discarded.

[0090] Embodiment Two

[0091] Embodiment Two of the present application provides a system for cervical liquid-based cell segmentation based on a watershed, comprising:

[0092] A preprocessing module configured to acquire a cervical liquid-based cell image, and perform grayscale, thresholding, and opening operation preprocessing operations on the cervical liquid-based cell image.

[0093] A first segmentation module configured to calculate foreground markers, background markers, and unknown regions in the image, and to complete preliminary coarse segmentation using a watershed method.

[0094] A second segmentation module configured to obtain all target contours of the coarse segmentation, and to calculate the area, perimeter, circularity, and solidity information of each target, and to divide the cells into under-segmented, over-segmented, and final targets using a Kmeans method.

[0095] A target optimization module configured to optimize over-segmented and under-segmented targets, and to add the optimized targets to the final targets.

[0096] The steps and methods involved in the above embodiment two system correspond to embodiment one, and the specific implementation can be seen in the relevant description of embodiment one.

[0097] Those skilled in the art should understand that the above modules or steps of the present application can be realized by a general computer device. Alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device and executed by a computing device, or they can be respectively manufactured into individual integrated circuit modules, or a plurality of modules or steps among them can be manufactured into a single integrated circuit module. The present application is not limited to any specific combination of hardware and software.

[0098] The above describes the specific embodiments of the present application in combination with the drawings, but is not a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the protection scope of the present application.

Claims

1. A watershed-based cervical liquid-based cell segmentation method, characterized in that, The method comprises the following steps: Step S1: Obtain a cervical liquid-based cell image, and perform grayscale, thresholding, and opening operation preprocessing operations on the cervical liquid-based cell image; Step S2: Calculate foreground markers, background markers, and unknown regions in the image, and use a watershed method to complete preliminary rough segmentation; Step S3: Obtain all target contours of rough segmentation, and calculate relevant information of each target, and use a Kmeans method to divide the cells into under-segmentation, over-segmentation, and final targets; Step S4: Optimize over-segmentation targets and under-segmentation targets, and add the optimized targets to the final targets; The optimization of the under-segmentation target specifically comprises: cropping the under-segmentation target region, calculating the coordinates of the circumscribed rectangle of each under-segmentation target, and cropping the under-segmentation target image from the original cervical cell image; eliminating the background of the under-segmentation region image, filling the non-over-segmentation part in the cropped over-segmentation rectangle image with a pixel value of (0, 0, 0), only retaining the over-segmentation target, and eliminating other targets that are cropped together in the image; performing grayscale processing on the under-segmentation region image, converting the RGB three-channel image into a single-channel grayscale image; performing secondary thresholding on the under-segmentation region image, converting the grayscale image into a binary image; performing a target searching operation on the binary image, judging whether the area of each target is in the interval [40, 1500], and if so, adding the target to the final targets; cropping the over-segmentation target region, calculating the coordinates of the circumscribed rectangle of each over-segmentation target, and widening the target surrounding region, and cropping the over-segmentation target image from the original cervical cell image; The grayscale processing on the over-segmentation region image converts the RGB three-channel image into a single-channel grayscale image; the thresholding processing on the grayscale image converts the grayscale image into a binary image; the dilation processing on the binary image causes the under-segmentation target boundary in the binary image to expand outward; if two objects are close to each other, the two objects will be connected together; the target searching operation on the binary image judges whether the area of each target is in the interval [40, 1500], and if so, the target is added to the final targets.

2. The cervical liquid-based cell segmentation method based on a watershed, according to claim 1, wherein the preprocessing operation specifically comprises the following steps: performing grayscale processing on the original cervical cell image to obtain a grayscale image; the original image uses an RGB color mode, and each pixel point is represented by three color variables of red, green, and blue; each RGB component of each pixel is assigned an intensity value in the range of 0-255; and the grayscale conversion is performed to convert the single-channel image into an image having only one gray value intensity component and a gray range of 0-255; performing binaryzation on the grayscale image to obtain a binary image; the grayscale image having pixel values of 0-255 is converted into a black-and-white image having only two values of 0 and 255; performing denoising processing on the binary image; the binary image is processed by using a morphological opening operation, i.e., first erosion and then dilation. The specific steps of calculating the foreground markers, the background markers, and the unknown regions in the image comprise:

3. The watershed-based cervical liquid-based cell segmentation method of claim 1, wherein, ​ The background marker of the image is obtained, morphological dilation processing is performed on the target of interest in the binary image, the boundary points of the binary object are expanded, all background points in contact with the object are merged into the object, the boundary of the target of interest of the image is expanded outward, and the background marker contains more true target of interest; The foreground marker of the image is obtained, a distance transform operation is performed on the binary image, the distance of each non-zero point in the image from the nearest zero point is calculated, the gray value matrix of the binary image is converted into a distance matrix to the nearest zero point, and the foreground marker of the image is obtained through binaryzation of the distance matrix; The unknown region marker of the image is obtained, and the unknown region marker matrix of the image is obtained by subtracting the foreground marker matrix from the background marker matrix.

4. The watershed-based cervival liquid-based cell segmentation method of claim 3, wherein, The specific steps of completing the preliminary rough segmentation using the watershed method include: marking the unknown region in the foreground marker as 0 pixel value through connected domain analysis; The target of interest is segmented from the original image using the watershed method, the contour of the foreground marker is transmitted as a seed, other pixel points on the image are judged according to the rules of the watershed algorithm, and the region attribution of each pixel point is demarcated to obtain all cell targets of interest and complete the initial rough segmentation.

5. The watershed-based cervival liquid-based cell segmentation method of claim 1, wherein, The relevant information of each target includes: area, perimeter, convex hull, circumscribed circle area, circular granularity, solidity, major axis, minor axis, solidity, ellipticity, width, height and eccentricity.

6. The watershed-based cervival liquid-based cell segmentation method of claim 1, wherein, The specific steps of dividing the cells into under-segmentation, over-segmentation and final targets using the Kmeans method are as follows: Randomly select 3 samples as initial cluster centers; For each sample in the data set, calculate the distance from it to the k cluster centers, and attribute it to the class corresponding to the cluster center with the smallest distance; For each cluster, recalculate the cluster center position; Repeat the iteration operation until the maximum number of iterations is reached or the cluster center position is unchanged.

7. The watershed-based cervical liquid-based cell segmentation method of claim 1, wherein, The under-segmented targets are optimized using secondary thresholding, and the over-segmented targets are optimized using dilation processing.

8. A system for cervix liquid-based cell segmentation based on watershed, characterized in that, It includes: A preprocessing module configured to obtain a cervical liquid-based cell image, perform grayscale, thresholding and opening operation preprocessing on the cervical liquid-based cell image; A first segmentation module configured to calculate the foreground marker, background marker and unknown region of the image, and complete preliminary rough segmentation using the watershed method; A second segmentation module configured to obtain the contour of all targets of rough segmentation, and calculate the area, perimeter, circular granularity and solidity information of each target, and divide the cells into under-segmentation, over-segmentation and final targets using the Kmeans method; A target optimization module configured to optimize the over-segmented targets and under-segmented targets, and add the optimized targets to the final targets; For optimization of under-segmentation targets, specifically including: cropping under-segmentation target regions, calculating the coordinates of the outer rectangle of each under-segmentation target, and cropping the under-segmentation target image from the original cervical cell image; eliminating the background of the under-segmentation region image, filling the non-over-segmentation part of the cropped over-segmentation rectangle image with pixel values of (0, 0, 0), only retaining the over-segmentation target, and eliminating other targets that are cropped together in the image; grayscale processing of the under-segmentation region image, converting the RGB three-channel image into a single-channel grayscale image; secondary thresholding of the under-segmentation region image, converting the grayscale image into a binary image; finding the target operation on the binary image, judging whether the area of each target is in the [40, 1500] interval, if so, adding the final target; cropping the over-segmentation target region, calculating the coordinates of the outer rectangle of each over-segmentation target, and widening the target surrounding area, cropping the over-segmentation target image from the original cervical cell image; Grayscale processing of the over-segmentation region image, converting the RGB three-channel image into a single-channel grayscale image; thresholding of the grayscale image, converting the grayscale image into a binary image; dilation processing of the binary image, expanding the under-segmentation target boundary in the binary image outward; if two objects are close together, they will be connected together; finding the target operation on the binary image, judging whether the area of each target is in the [40, 1500] interval, if so, adding the final target.

Images