An intelligent enhancement method for fundus vascular images

Abstract

The present invention relates to the technical field of fundus vascular image enhancement, and specifically relates to an intelligent enhancement method for fundus vascular images. First, the present invention acquires a fundus vascular image and obtains the vascular region; further, according to the change characteristics of the number of pixel points before and after erosion, the erosion index of each vascular region is obtained; further, according to the difference characteristics of the erosion indexes of the vascular regions within the preset neighborhood window of the target region, the small vessel coefficient of the target region is obtained, and the potential small vessel region is obtained; further, according to the difference characteristics between the erosion index of the potential small vessel region and the erosion index of the adjacent vascular region, and the distribution characteristics of all erosion indexes, combined with the small vessel coefficient, the authenticity coefficient of the potential small vessel region is obtained, and the small vessel region is obtained; finally, according to the erosion index and the authenticity coefficient of the small vessel region, the enhancement coefficient of the small vessel region is obtained, and image enhancement is performed on the fundus vascular image.

Description

Technical Field

[0001] The present invention relates to the technical field of fundus vascular image enhancement, and particularly to an intelligent enhancement method for fundus vascular images. Background Art

[0002] With the change of people's lifestyles, various electronic products have penetrated into all aspects of life. Various unhealthy eye - using habits and excessive eye fatigue have caused more and more patients with eye diseases. Various eye diseases will cause varying degrees of changes to the fundus structure. Fundus images are the most direct and effective basis for doctors to diagnose eye diseases. The retina at the bottom of the human eye contains a large number of tiny blood vessels. The trends, diameters, and shape structures of these micro - blood vessels are related to systemic diseases such as eye diseases, cardiovascular and cerebrovascular diseases, and hypertension.

[0003] During the existing process of enhancing fundus vascular images, the vascular skeleton is extracted through morphological erosion, the vascular break points are obtained to determine the parts with smaller vascular diameters, and the areas with smaller vascular diameters are enhanced. However, problems such as excessive reduction, inaccurate break - point judgment, and noise interference may occur during the extraction of the vascular skeleton, resulting in inaccurate judgment of tiny blood vessels and affecting the image quality. In addition, the macula area may also interfere with the judgment of tiny blood vessels and affect the enhancement effect. Summary of the Invention

[0004] In order to solve the technical problems of inaccurate positioning of tiny blood vessel areas and unsatisfactory image enhancement effect in the existing technology, the purpose of the present invention is to provide an intelligent enhancement method for fundus vascular images. The specific technical solutions adopted are as follows:

[0005] An intelligent enhancement method for fundus vascular images, the method includes:

[0006] Obtain a fundus vascular image and obtain the vascular area; perform morphological erosion on each of the vascular areas;

[0007] According to the change characteristics of the number of pixel points before and after erosion, obtain the erosion index of each vascular area; record from the root of the blood vessel to any end point as a blood - vessel branch; select any one of the blood - vessel branches as the target branch; select any one of the vascular areas on the target branch as the target area; according to the difference characteristics of the erosion indexes of the vascular areas within the preset neighborhood window of the target area, obtain the tiny blood vessel coefficient of the target area; obtain the potential tiny blood vessel area according to the tiny blood vessel coefficient;

[0008] Based on the difference characteristics between the corrosion index of the potential small blood vessel region and that of the adjacent blood vessel region, as well as the distribution characteristics of all the corrosion indices, and in combination with the small blood vessel coefficient, obtain the authenticity coefficient of the potential small blood vessel region; based on the authenticity coefficient, obtain the small blood vessel region; based on the corrosion index and the authenticity coefficient of each small blood vessel region, obtain the enhancement coefficient of each small blood vessel region;

[0009] Perform image enhancement on the fundus blood vessel image according to the enhancement coefficient.

[0010] Further, the method for obtaining the corrosion index includes:

[0011] In each blood vessel region, take the ratio of the change value of the number of pixel points before and after corrosion to the number of pixel points before corrosion as the corrosion index of each blood vessel region.

[0012] Further, the method for obtaining the small blood vessel coefficient includes:

[0013] The preset neighborhood window is centered on the target region;

[0014] Based on the difference characteristics between the corrosion index of the target region and that of the two adjacent blood vessel regions on both sides, and in combination with the maximum difference between the corrosion indices among all the blood vessel regions within the preset neighborhood window, obtain the small blood vessel coefficient of the target region; the difference characteristics between the corrosion index of the target region and that of the two adjacent blood vessel regions on both sides are positively correlated with the small blood vessel coefficient; the maximum difference is negatively correlated with the small blood vessel coefficient.

[0015] Further, the method for obtaining the potential small blood vessel region includes:

[0016] When the small blood vessel coefficient is greater than or equal to the first preset threshold, mark the corresponding blood vessel region as a potential small blood vessel region.

[0017] Further, the method for obtaining the authenticity coefficient includes:

[0018] Taking the root of the blood vessel as the starting point, every time a branch point is passed, the level of the blood vessel region increases by 1, and perform level division on each blood vessel region; cluster the corrosion indices of all the blood vessel regions in the fundus blood vessel image; based on the difference characteristics of the corrosion indices of all the blood vessel regions in each cluster, and in combination with the difference characteristics of the levels of all the blood vessel regions, obtain the macula cluster;

[0019] Based on the difference between the corrosion index of each potential small blood vessel region and the corrosion indices of the two adjacent blood vessel regions on both sides, combined with the distance between the corrosion index of the potential small blood vessel region and the center of the macula clustering cluster, and the small blood vessel coefficient of the potential small blood vessel region, the authenticity coefficient of each potential small blood vessel region is obtained; the difference between the corrosion index of the potential small blood vessel region and the corrosion indices of the two adjacent blood vessel regions on both sides is negatively correlated with the authenticity coefficient; the distance between the corrosion index of the potential small blood vessel region and the center of the macula clustering cluster, and the small blood vessel coefficient, are both positively correlated with the authenticity coefficient.

[0020] Further, the method for obtaining the macula clustering cluster includes:

[0021] In each clustering cluster, the ratio of the variance of all the blood vessel region grades to the variance of the corrosion index is used as the macula possibility of each clustering cluster; the clustering cluster with the largest macula possibility is selected as the macula clustering cluster.

[0022] Further, the method for obtaining the small blood vessel region includes:

[0023] When the authenticity coefficient of the potential small blood vessel region is greater than a second preset threshold, the corresponding potential small blood vessel region is marked as a small blood vessel region.

[0024] Further, the method for obtaining the enhancement coefficient includes:

[0025] After normalizing the ratio of the authenticity coefficient of the small blood vessel region to the corrosion index, it is used as the enhancement coefficient corresponding to the small blood vessel region.

[0026] Further, the method for performing image enhancement on the fundus blood vessel image according to the enhancement coefficient includes:

[0027] The sum value of the enhancement coefficient and the constant 1 is used as the enhancement parameter;

[0028] The product of the contrast of the small blood vessel region and the enhancement parameter is used as the corrected contrast corresponding to the small blood vessel region; based on the corrected contrasts of all the small blood vessel regions, image enhancement is performed on the fundus blood vessel image.

[0029] Further, the method for obtaining the blood vessel region includes:

[0030] Obtain the binary image of the fundus blood vessel image, extract the blood vessel skeleton of the binary image, and extract endpoints, branch points, and roots based on feature point detection;

[0031] Vascular pixel points between adjacent vascular branch points, between an endpoint and an adjacent vascular branch point, and between a root and an adjacent vascular branch point are denoted as one of the said vascular regions.

[0032] The present invention has the following beneficial effects:

[0033] The present invention first obtains a fundus vascular image and obtains a vascular region to provide an analysis basis; further performs morphological erosion on each vascular region to facilitate optimization based on morphological erosion, so as to more accurately locate fine blood vessels and reduce macular interference; further obtains the erosion index of each vascular region to quantify the change in pixel points of erosion and at the same time reflect the thickness of blood vessels, providing a basis for locating fine blood vessels; further obtains the fine blood vessel coefficient of the target region according to the difference characteristics of the erosion indexes of vascular regions within the preset neighborhood window of the target region, obtains a potential fine blood vessel region, analyzes the possibility that the target blood vessel is a fine blood vessel from a local perspective, initially screens the vascular region, and provides a basis for finally accurately locating fine blood vessels and performing intelligent enhancement on the image; further, based on the fine blood vessel coefficient, evaluates the possibility that a potential fine blood vessel belongs to a real fine blood vessel from a global perspective, obtains the authenticity coefficient of the potential fine blood vessel region, and provides a basis for accurately locating the fine blood vessel region; further, according to the authenticity coefficient, obtains the fine blood vessel region, accurately locates the fine blood vessel region, and provides a basis for performing intelligent enhancement on the fundus vascular image; finally, according to the erosion index and authenticity coefficient of each fine blood vessel region, obtains the enhancement coefficient of each fine blood vessel region, performs image enhancement on the fundus vascular image, and improves the details of the fine blood vessel region in the fundus vascular image. The present invention utilizes the characteristics that blood vessels with different diameters have different changes after morphological erosion, and utilizes the characteristics that the erosion indexes between macular regions are similar, screens out potential fine blood vessel regions and evaluates their authenticity, finally accurately locates the fine blood vessel regions and performs intelligent enhancement on the image, improving the image details. Description of the Drawings

[0034] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0035] Figure 1 It is a flowchart of a method for intelligent enhancement of a fundus vascular image provided by an embodiment of the present invention;

[0036] Figure 2 It is a fundus vascular image provided by an embodiment of the present invention;

[0037] Figure 3An image after morphological erosion of a fundus vascular image provided by an embodiment of the present invention;

[0038] Figure 4 A schematic diagram of grade division provided by an embodiment of the present invention. Detailed implementation manners

[0039] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following combines the accompanying drawings and preferred embodiments to specifically describe a method for intelligent enhancement of fundus vascular images proposed according to the present invention, including its specific implementation manners, structures, features and effects in detail. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0041] The following specifically describes the specific solution of a method for intelligent enhancement of fundus vascular images provided by the present invention with reference to the accompanying drawings.

[0042] Please refer to Figure 1 , which shows a flowchart of a method for intelligent enhancement of fundus vascular images provided by an embodiment of the present invention, specifically including:

[0043] Step S1: Obtain a fundus vascular image and obtain the vascular region; perform morphological erosion on each vascular region.

[0044] A large number of tiny blood vessels are contained in the retina at the bottom of the human eyeball. The trends, diameters, and shape structures of these microvessels are all related to systemic diseases such as eye diseases, cardiovascular and cerebrovascular diseases, and hypertension. For example, diabetic retinopathy usually manifests as damage, occlusion, or dilation of tiny blood vessels; hypertensive retinopathy may cause sclerosis or stenosis of retinal arterioles, and these changes often first appear in smaller blood vessels. The small blood vessels in fundus vascular images are usually relatively subtle and complex in shape, and the blood flow velocity is slow, which makes them may be difficult to detect or clearly observe in fundus images, especially in the early stages of some diseases (such as diabetic retinopathy, glaucoma, hypertensive retinopathy, etc.). The tiny changes in small blood vessels may become the key to early diagnosis. Therefore, ensuring the image clarity and contrast of these small blood vessels is the key to improving the diagnostic accuracy, and it is necessary to enhance the images of these small blood vessels.

[0045] The embodiment of the present invention first obtains a fundus vascular image to provide an analysis basis; please refer to Figure 2, which shows a fundus vascular image provided by an embodiment of the present invention. To determine the fine blood vessels in the fundus vascular image, first, the vascular region is obtained to determine the analysis range.

[0046] Preferably, in an embodiment of the present invention, threshold segmentation is performed on the fundus vascular image to obtain the vascular region; by performing threshold segmentation on Figure 2 a binary image is obtained, and then the vascular skeleton is extracted by using the skeletonization technique. Then, based on feature point detection, the end points and branch points are extracted, where the end point is a pixel point with only one adjacent pixel point, and the branch point is a pixel point with three or more adjacent pixel points. The branch point with the largest diameter is used as the root, and the end points, roots, and branch points are extracted. The vascular pixel points between adjacent vascular branch points, between the end point and adjacent vascular branch points, and between the root and adjacent vascular branch points are recorded as a vascular region.

[0047] It should be noted that threshold segmentation to obtain a binary image, the skeletonization technique, and end point detection are all existing technologies. Specifically, the Otsu method or the Frangi filter can be used to obtain the binary image, and the iterative shrinkage method or Voronoi skeleton can be used for skeletonization; in other embodiments of the present invention, the implementer can also train a convolutional neural network by manually annotating samples and use the neural network to obtain the vascular region in the fundus vascular image, which will not be elaborated here.

[0048] Considering that the existing morphological erosion may cause problems such as excessive reduction, inaccurate judgment of break points, and noise interference when extracting the vascular skeleton and locating fine blood vessels, resulting in inaccurate judgment of fine blood vessels and affecting the image quality; in addition, the macular area may also interfere with the judgment of fine blood vessels and affect the enhancement effect. Therefore, based on morphological erosion, optimization is carried out to more accurately locate fine blood vessels and reduce macular interference. Therefore, morphological erosion is performed on each vascular region. Please refer to Figure 3 , which shows an image after morphological erosion of a fundus vascular image provided by an embodiment of the present invention.

[0049] It should be noted that the morphological erosion method is already a well-known technical means in the art and will not be elaborated here.

[0050] Step S2: According to the change characteristics of the number of pixel points before and after erosion, obtain the erosion index of each vascular region; record from the vascular root to any end point as a vascular branch; select any vascular branch as the target branch; select any vascular region on the target branch as the target region; according to the difference characteristics of the erosion indexes of the vascular regions within the preset neighborhood window of the target region, obtain the fine blood vessel coefficient of the target region; obtain the potential fine blood vessel region according to the fine blood vessel coefficient.

[0051] Considering that for blood vessels with a larger diameter or thicker, the number of pixel points in the blood vessel area changes significantly after morphological erosion; while for blood vessel areas with thinner blood vessels, due to the thinness of the blood vessels themselves, the number of pixel points that can be removed by erosion is limited. Therefore, there are differences in the change of the number of pixel points before and after erosion between thin blood vessels and thicker blood vessels. Therefore, according to the change characteristics of the number of pixel points before and after erosion, the erosion index of each blood vessel area is obtained to quantify the change of pixel points eroded, providing a basis for determining the thin blood vessel area.

[0052] Preferably, in an embodiment of the present invention, considering that the larger the ratio of the change value of the number of pixel points before and after erosion to the number of pixel points before erosion, the greater the impact of erosion on the blood vessel area, the larger the erosion index, and the more likely it is a thicker blood vessel; in each blood vessel area, the ratio of the change value of the number of pixel points before and after erosion to the number of pixel points before erosion is used as the erosion index of each blood vessel area.

[0053] It should be noted that before and after erosion refers to before and after morphological erosion, before erosion is before morphological erosion, and after erosion is after morphological erosion; the change value of the number of pixel points before and after erosion is the absolute value of the difference between the number of pixel points before erosion and the number of pixel points after erosion; in another embodiment of the present invention, in each blood vessel area, the change value of the number of pixel points before and after erosion is linearly normalized and used as the erosion index of each blood vessel area.

[0054] In the fundus vascular skeleton, since the central artery starting from the optic disc branches, the blood vessels will enter different layers of the retina through successive branching and finally form a tiny capillary network. Therefore, the diameter of the fundus blood vessels shows a regular change from thick to thin. However, in some special cases (such as blood vessel branching, pathological changes or physiological regulation, etc.), there may be a short-term contraction of the retinal blood vessels. Therefore, during the process of the blood vessel diameter changing from thick to thin, there may be a situation where the local area becomes abnormally thin.

[0055] Since the erosion indices of blood vessel areas with different thicknesses are different, the possibility of thin blood vessels can be analyzed through the change of the erosion index of the blood vessel area on the blood vessel branch. Denote from the root of the blood vessel to any endpoint as a blood vessel branch; select any blood vessel branch as the target branch; select any blood vessel area on the target branch as the target area; considering that when the blood vessel shows local abnormal thinning, the erosion index is quite different from the erosion indices of other blood vessel areas in the local neighborhood, and the smoothness of the change of the erosion index in the local neighborhood is affected. Therefore, according to the difference characteristics of the erosion indices of the blood vessel areas within the preset neighborhood window of the target area, the thin blood vessel coefficient of the target area is obtained, and the possibility of the target blood vessel being a thin blood vessel is analyzed from a local perspective, providing a basis for finally accurately locating the thin blood vessels and performing intelligent enhancement on the image.

[0056] Preferably, in an embodiment of the present invention, considering that the greater the difference in the corrosion index between the target region and the blood vessel regions on both sides within the preset neighborhood window, the worse the smoothness of the change in the corrosion index within the local neighborhood, the less smooth the process of the blood vessel changing from thick to thin, indicating that the target region is more likely to be a small blood vessel region and the greater the small blood vessel coefficient; at the same time, the greater the difference between the target region and the blood vessel regions on both sides relative to the maximum difference within the preset neighborhood window, the more prominent the change in the corrosion index of the target region within the local neighborhood and the more obvious the small blood vessel characteristics. Based on this:

[0057] According to the difference characteristics between the corrosion index of the target region and the corrosion indices of the adjacent blood vessel regions on both sides, combined with the maximum difference in the corrosion indices between all blood vessel regions within the preset neighborhood window, obtain the small blood vessel coefficient of the target region; the difference characteristics between the corrosion index of the target region and the corrosion indices of the adjacent blood vessel regions on both sides are positively correlated with the small blood vessel coefficient; the maximum difference is negatively correlated with the small blood vessel coefficient.

[0058] As an example, the length of the preset neighborhood window is 5, that is, it contains 5 blood vessel regions. The target region is the center of the preset neighborhood window, and there are two adjacent blood vessel regions on each side; within the preset neighborhood window of the target region, take the blood vessel region on either side as the left adjacent blood vessel region and the blood vessel region on the other side as the right adjacent blood vessel region; the calculation formula for the small blood vessel coefficient includes:

[0059] ;

[0060] where, i represents the serial number of the target region; represents the small blood vessel coefficient of the i-th target region; represents the corrosion index of the i-th target region; represents the average value of the corrosion indices of the left adjacent blood vessel region of the i-th target region; represents the average value of the corrosion indices of the right adjacent blood vessel region of the i-th target region; represents the maximum value of the corrosion indices of all blood vessel regions within the preset neighborhood window of the i-th target region; represents the minimum value of the corrosion indices of all blood vessel regions within the preset neighborhood window of the i-th target region.

[0061] In the calculation formula of the small blood vessel coefficient, the corrosion index of the adjacent blood vessel region on one side is represented by the average value; the difference characteristics between the corrosion index of the target blood vessel and the corrosion index of the adjacent region on one side are represented by the absolute value of the difference, and then the average value of the corresponding absolute values of the differences on both sides is obtained to represent the difference characteristics between the corrosion index of the target region and the corrosion indices of the adjacent blood vessel regions on both sides; The larger it is, the greater the difference in the corrosion index between the target region and the vascular regions on both sides within the preset neighborhood window, reflecting that the process of the blood vessel changing from thick to thin is less smooth, indicating that the target region is more likely to be a small blood vessel region, and the small blood vessel coefficient is larger; The larger it is, the greater the difference between the target region and the vascular regions on both sides relative to the maximum difference within the preset neighborhood window, indicating that the change in the corrosion index of the target region is more prominent within the local neighborhood, the small blood vessel characteristics are more obvious, and the small blood vessel coefficient is larger.

[0062] It should be noted that when there are insufficient adjacent vascular regions in the target region, such as only 1 on one side, the corrosion index of the vascular region on this side is used to represent the corrosion index of the adjacent vascular region on the corresponding side at this time; for another example, when there is only one side, only the difference on one side is compared, and the average value is no longer calculated, such as .

[0063] It should be noted that considering that multiple blood vessel branches may pass through the same vascular region, in one embodiment of the present invention, the average value of the small blood vessel coefficients of all blood vessel branches corresponding to the vascular region is used as the final small blood vessel coefficient; in another embodiment of the present invention, the implementer can also select the longest blood vessel branch flowing through the same vascular region as the attributed blood vessel branch of this vascular region, and only calculate the small blood vessel coefficient in the longest blood vessel branch.

[0064] The small blood vessel coefficient measures the possibility that the target blood vessel is a small blood vessel from a local perspective. Therefore, potential small blood vessel regions are obtained according to the small blood vessel coefficient, the vascular regions are preliminarily screened, and the analysis range is narrowed.

[0065] Preferably, in one embodiment of the present invention, considering that the larger the small blood vessel coefficient, the more likely the vascular region is a small blood vessel, so when the small blood vessel coefficient is greater than or equal to the first preset threshold, the corresponding vascular region is marked as a potential small blood vessel region.

[0066] As an example, the first preset threshold is 0.5.

[0067] Step S3: According to the difference characteristics between the corrosion index of the potential small blood vessel region and the corrosion index of the adjacent vascular region, as well as the distribution characteristics of all corrosion indexes, combined with the small blood vessel coefficient, obtain the authenticity coefficient of the potential small blood vessel region; obtain the small blood vessel region according to the authenticity coefficient; obtain the enhancement coefficient of each small blood vessel region according to the corrosion index and authenticity coefficient of each small blood vessel region.

[0068] Considering that the fundus vascular image contains a macula region, when performing morphological erosion, the macula region will also be eroded, and the macula region may be identified as the breakpoint of a small blood vessel or misidentified as a small blood vessel region. Therefore, it is also necessary to further analyze the potential small blood vessel region to exclude the interference of the macula region.

[0069] Considering that when different macular regions are eroded, the erosion indices are relatively similar, while for blood vessel regions with different thicknesses in blood vessel branches, there are significant differences in erosion indices. Therefore, taking advantage of this feature, in the blood vessel branches where potential tiny blood vessel regions are located, the tiny blood vessel coefficient is further corrected according to the distribution characteristics of all erosion indices. Also, considering that when tiny blood vessels appear in blood vessel branches, there are significant differences in erosion indices between the tiny blood vessels and adjacent blood vessel regions. Therefore, based on the difference characteristics between the erosion indices of potential tiny blood vessel regions and adjacent blood vessel regions, the authenticity coefficient of potential tiny blood vessel regions is obtained, further evaluating the possibility that potential tiny blood vessels belong to real tiny blood vessels from a global perspective, providing a basis for accurately locating tiny blood vessel regions.

[0070] Preferably, in an embodiment of the present invention, considering that from the root to the end point, blood vessels gradually become thinner, and the change is more obvious before and after the branch point. Therefore, starting from the root of the blood vessel, every time a branch point is passed, the blood vessel region level increases by 1, and each blood vessel region is classified by level. Please refer to Figure 4 , which shows a schematic diagram of level classification provided by an embodiment of the present invention, Figure 4 In which each line represents a blood vessel region, and the number represents the level of the blood vessel region. The level is used to represent the distance between the blood vessel region and the root, and at the same time represents the thickness of the blood vessel. The higher the level, the farther the distance from the root and the thinner the blood vessel;

[0071] Considering that the erosion indices of macular regions are relatively similar, that is, the differences in erosion indices of macular regions at different levels are relatively small. Therefore, the erosion indices of all blood vessel regions in the fundus blood vessel image are clustered. According to the difference characteristics of the erosion indices of all blood vessel regions in each clustering cluster, combined with the difference characteristics of the levels of all blood vessel regions, the macular clustering cluster is obtained, and the clustering cluster where the macular region is located is determined;

[0072] Considering that the distance between real tiny blood vessel regions and the macular clustering cluster is relatively far, and due to the abnormal thinning of blood vessels, the erosion index is very likely to be lower than that of the adjacent blood vessel regions on both sides. At the same time, the tiny blood vessel coefficient measures the possibility that a blood vessel region is a tiny blood vessel region from a local perspective. The larger the tiny blood vessel coefficient, the more likely the blood vessel region is a tiny blood vessel. Based on this:

[0073] According to the difference between the corrosion index of each potential small blood vessel region and the corrosion indices of the two adjacent blood vessel regions on both sides, combined with the distance between the corrosion index of the potential small blood vessel region and the center of the macular clustering cluster, and the small blood vessel coefficient of the potential small blood vessel region, the authenticity coefficient of each potential small blood vessel region is obtained; the difference between the corrosion index of the potential small blood vessel region and the corrosion indices of the two adjacent blood vessel regions on both sides is negatively correlated with the authenticity coefficient; the distance between the corrosion index of the potential small blood vessel region and the center of the macular clustering cluster, and the small blood vessel coefficient, are both positively correlated with the authenticity coefficient.

[0074] The calculation formula for the authenticity coefficient includes:

[0075] ;

[0076] where k represents the serial number of the potential small blood vessel; represents the authenticity coefficient of the k-th potential small blood vessel; represents the linear normalization function; represents the small blood vessel coefficient of the k-th potential small blood vessel; represents the distance between the corrosion index of the k-th potential small blood vessel and the center of the macular clustering cluster; represents the exponential function with the natural constant e as the base; represents the corrosion index of the k-th potential small blood vessel; represents the corrosion index of the blood vessel region with the smallest grade adjacent to the k-th potential small blood vessel; represents the corrosion index of the blood vessel region with the largest grade adjacent to the k-th potential small blood vessel.

[0077] In the calculation formula for the authenticity coefficient, through the negative correlation mapping function the corrosion index difference is negatively correlated and mapped, where x represents the independent variable; in and the smaller they are, the smaller the corrosion index of the potential small blood vessel region is relative to the corrosion indices of the two adjacent blood vessel regions on both sides, the more obvious the feature of abnormal thinning of the potential small blood vessel region is, and the more likely it is to be a real small blood vessel, and the larger the real coefficient; the larger it is, the greater the difference between the potential small blood vessel region and the macular clustering cluster in the macular region, the less likely it is to be the macular region, and the larger the authenticity coefficient; the larger it is, from a local perspective, the more likely the blood vessel region is to be a small blood vessel, and the larger the authenticity coefficient.

[0078] It should be noted that in an embodiment of the present invention, the k-means clustering algorithm is used for clustering. The clustering center is the mean value of all corrosion indexes in the clustering cluster, and the distance from the corrosion index to the clustering cluster center is the absolute value of the difference between the corrosion index and the clustering center. When there is only one adjacent blood vessel area for the potential small blood vessel, only the difference corresponding to one adjacent blood vessel is calculated without taking the average. If not When .

[0079] In other embodiments of the present invention, the implementer can also use other clustering algorithms such as DBSCAN clustering. Both of them are already existing technologies and will not be elaborated here.

[0080] Preferably, in an embodiment of the present invention, considering that the more chaotic the grades of the blood vessel areas within the clustering cluster are, and at the same time the more concentrated the corrosion indexes are, it indicates that the clustering cluster more conforms to the characteristics of the macula area with different grades and similar corrosion indexes. Also considering that the larger the variance is, the greater the data fluctuation and chaos are. Therefore, in each clustering cluster, the ratio of the variance of all blood vessel area grades to the variance of the corrosion indexes is used as the macula possibility of each clustering cluster. The clustering cluster with the largest macula possibility is selected as the macula clustering cluster.

[0081] It should be noted that usually there are certain differences in the corrosion indexes of different blood vessel areas, that is, the variance of the corrosion indexes is not zero. The implementer can also use a preset positive parameter for dividing by zero, such as 0.01, take the variance of all blood vessel area grades as the numerator, and take the sum of the variance of the corrosion indexes and 0.01 as the denominator, and the fractional ratio is used as the macula possibility of each clustering cluster.

[0082] It should be noted that when the potential small blood vessel area is close to the branch point, there may be multiple adjacent blood vessel areas on one side. For example, Figure 4 for the middle blood vessel area with grade 2 in, there are 3 adjacent blood vessel areas with grade 3 on one side. At this time, the implementer can stipulate to calculate the average value of the corrosion indexes of the 3 adjacent blood vessel areas as the corrosion index of the blood vessel area with the largest grade adjacent to the potential small blood vessel. It can also be stipulated to select the blood vessel area on the longest blood vessel branch, that is, select the corrosion index of the rightmost blood vessel area as the corrosion index of the blood vessel area with the largest grade adjacent to the potential small blood vessel.

[0083] Based on the small blood vessel coefficient evaluated locally, the authenticity of the potential small blood vessel area is further evaluated from a global perspective to obtain the authenticity coefficient. Then, according to the authenticity coefficient, the small blood vessel area is obtained to accurately locate the small blood vessel area, providing a basis for the intelligent enhancement of the fundus blood vessel image.

[0084] Preferably, in an embodiment of the present invention, considering that the larger the authenticity coefficient is, the more likely the potential small blood vessel area is a real small blood vessel area. Therefore, when the authenticity coefficient of the potential small blood vessel area is greater than the second preset threshold, the corresponding potential small blood vessel area is marked as a small blood vessel area.

[0085] As an example, the second preset threshold is 0.5.

[0086] The corrosion index reflects the change in the number of pixel points of the small blood vessels before and after corrosion, reflecting the thickness of the small blood vessels; the authenticity coefficient characterizes the authenticity of the small blood vessels. Therefore, according to the corrosion index and authenticity coefficient of each small blood vessel area, the enhancement coefficient of each small blood vessel area is obtained, providing an enhancement basis for accurately enhancing the small blood vessels.

[0087] Preferably, in an embodiment of the present invention, considering that the larger the authenticity coefficient is, the more real the small blood vessels are; the smaller the corrosion index is, the thinner the small blood vessels are. The thinner and more real the small blood vessels are, the more they need to be enhanced. Therefore, after normalizing the ratio of the authenticity coefficient of the small blood vessel area to the corrosion index, it is used as the enhancement coefficient of the corresponding small blood vessel area.

[0088] It should be noted that a linear normalization method is specifically adopted; in other embodiments of the present invention, the implementer can also perform a negative correlation mapping on the corrosion index through a negative correlation mapping function such as exp(-x), and after linearly normalizing the product of the negative correlation mapping value and the authenticity coefficient, the normalization result is used as the enhancement coefficient of the corresponding small blood vessel area.

[0089] Step S4: Perform image enhancement on the fundus blood vessel image according to the enhancement coefficient.

[0090] The enhancement coefficient characterizes the degree of enhancement of the small blood vessels. After determining the enhancement coefficients of all small blood vessels, the fundus blood vessel image can be enhanced according to the enhancement coefficient, improving the details of the small blood vessel areas in the fundus blood vessel image, and thus providing more accurate image data for clinical diagnosis and prevention of related diseases.

[0091] Preferably, in an embodiment of the present invention, the sum value of the enhancement coefficient and the constant 1 is used as the enhancement parameter;

[0092] The product of the contrast of the small blood vessel area and the enhancement parameter is used as the corrected contrast of the corresponding small blood vessel area; based on the corrected contrasts of all small blood vessel areas, the fundus blood vessel image is enhanced.

[0093] In other embodiments of the present invention, the implementer can also map the enhancement coefficient to other positive intervals, such as mapping to [0, 2]. At this time, the enhancement parameter is [1, 3], and the maximum corrected contrast is 3 times the original contrast.

[0094] It should be noted that obtaining the contrast and the contrast enhancement method are already well-known technical means to those skilled in the art. In an embodiment of the present invention, a local comparison algorithm can be used to achieve local contrast enhancement; in other embodiments of the present invention, the implementer can obtain the average gray value of the small blood vessel area as the foreground gray value, take other areas in the fundus blood vessel image as the background area, obtain the average gray value of the background area as the background gray value, and use the ratio of the difference between the foreground gray value and the background gray value to the sum value of the foreground gray value and the background gray value as the contrast; after calculating the corrected contrast, linearly transform the pixel values of the pixel points in the small blood vessel area, which will not be elaborated here.

[0095] In other embodiments of the present invention, a generative adversarial network can also be used to enhance the texture of the small blood vessel area. When training the model, the enhancement coefficient is combined as a guiding parameter to focus on the details of the small blood vessel area. The generative adversarial network is already prior art and will not be elaborated.

[0096] In summary, in view of the technical problems of inaccurate positioning of small blood vessel areas and unsatisfactory image enhancement effect in the prior art, the present invention proposes an intelligent enhancement method for fundus blood vessel images. The present invention first obtains a fundus blood vessel image and obtains the blood vessel area; further obtains the erosion index of each blood vessel area according to the change characteristics of the number of pixel points before and after erosion; further obtains the small blood vessel coefficient of the target area according to the difference characteristics of the erosion indexes of the blood vessel areas within the preset neighborhood window of the target area, and obtains the potential small blood vessel area; further obtains the authenticity coefficient of the potential small blood vessel area according to the difference characteristics between the erosion index of the potential small blood vessel area and the erosion index of the adjacent blood vessel area, and the distribution characteristics of all erosion indexes, and combines the small blood vessel coefficient to obtain the small blood vessel area; finally, according to the erosion index and authenticity coefficient of each small blood vessel area, obtains the enhancement coefficient of each small blood vessel area, and performs image enhancement on the fundus blood vessel image. The present invention utilizes the characteristics that blood vessels with different diameters have different changes after morphological erosion, and the characteristics that the erosion indexes between macular areas are similar, screens out potential small blood vessel areas and evaluates their authenticity, and finally accurately locates the small blood vessel areas and performs intelligent enhancement on the image to improve the image details.

[0097] It should be noted that: the above order of the embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0098] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized respectively.

Claims

1. A method for intelligent enhancement of fundus vascular images, characterized in that: The method comprises: Acquire fundus vascular images and vascular regions; perform morphological corrosion on each of the vascular regions; According to the change characteristics of the number of pixel points before and after corrosion, the corrosion index of each blood vessel area is obtained; from the blood vessel root to any end point is recorded as a blood vessel branch; any blood vessel branch is selected as a target branch; any blood vessel area on the target branch is selected as a target area; according to the difference characteristics of the corrosion index of the blood vessel area in the preset neighborhood window of the target area, the small blood vessel coefficient of the target area is obtained; according to the small blood vessel coefficient, the potential small blood vessel area is obtained; According to the difference characteristics between the corrosion index of the potential small blood vessel area and the corrosion index of the adjacent blood vessel area, and the distribution characteristics of all the corrosion indexes, combined with the small blood vessel coefficient, the authenticity coefficient of the potential small blood vessel area is obtained; according to the authenticity coefficient, the small blood vessel area is obtained; according to the corrosion index and the authenticity coefficient of each small blood vessel area, the enhancement coefficient of each small blood vessel area is obtained; Performing image enhancement on the fundus vascular image according to the enhancement coefficient; The method for obtaining the authenticity coefficient includes: Taking the root of the blood vessel as the starting point, the blood vessel region level increases by 1 each time a branch point is passed, and each blood vessel region is graded; the corrosion indicators of all the blood vessel regions in the fundus blood vessel image are clustered; according to the difference characteristics of the corrosion indicators of all the blood vessel regions in each cluster, combined with the difference characteristics of the levels of all the blood vessel regions, a macular cluster is obtained; According to the difference between the corrosion index of each potential small blood vessel area and the corrosion index of the adjacent blood vessel areas on both sides, combined with the distance between the corrosion index of the potential small blood vessel area and the center of the macula cluster, and the small blood vessel coefficient of the potential small blood vessel area, the authenticity coefficient of each potential small blood vessel area is obtained; the difference between the corrosion index of the potential small blood vessel area and the corrosion index of the adjacent blood vessel areas on both sides is negatively correlated with the authenticity coefficient; the distance between the corrosion index of the potential small blood vessel area and the center of the macula cluster, and the small blood vessel coefficient are both positively correlated with the authenticity coefficient; The method for obtaining the macular cluster includes: In each of the clusters, the ratio of the variance of all the vascular area levels to the variance of the corrosion index is used as the macular possibility of each cluster; and the cluster with the largest macular possibility is selected as the macular cluster.

2. The method for intelligent enhancement of fundus vascular images according to claim 1, characterized in that: The method for obtaining the corrosion index includes: In each of the blood vessel regions, the ratio of the change in the number of pixel points before and after corrosion to the number of pixel points before corrosion is used as a corrosion index for each of the blood vessel regions.

3. The method for intelligent enhancement of fundus vascular images according to claim 1, characterized in that: The method for obtaining the small blood vessel coefficient includes: The preset neighborhood window is centered on the target area; According to the difference characteristics between the corrosion index of the target area and the corrosion index of the adjacent blood vessel areas on both sides, combined with the maximum difference of the corrosion index between all the blood vessel areas in the preset neighborhood window, the small blood vessel coefficient of the target area is obtained; the difference characteristics between the corrosion index of the target area and the corrosion index of the adjacent blood vessel areas on both sides are positively correlated with the small blood vessel coefficient; the maximum difference is negatively correlated with the small blood vessel coefficient.

4. The method for intelligent enhancement of fundus vascular images according to claim 3, characterized in that: The method for obtaining the potential small blood vessel area comprises: When the small blood vessel coefficient is greater than or equal to a first preset threshold, the corresponding blood vessel region is marked as a potential small blood vessel region.

5. The method for intelligent enhancement of fundus vascular images according to claim 1, characterized in that: The method for obtaining the small blood vessel region comprises: When the authenticity coefficient of the potential small blood vessel region is greater than a second preset threshold, the corresponding potential small blood vessel region is marked as a small blood vessel region.

6. The method for intelligent enhancement of fundus vascular images according to claim 1, characterized in that: The method for obtaining the enhancement coefficient includes: The ratio of the authenticity coefficient of the small blood vessel area to the corrosion index is normalized to be used as the enhancement coefficient corresponding to the small blood vessel area.

7. The method for intelligent enhancement of fundus vascular images according to claim 6, characterized in that: The method for performing image enhancement on the fundus vascular image according to the enhancement coefficient comprises: The sum of the enhancement coefficient and a constant 1 is used as an enhancement parameter; The product of the contrast of the small blood vessel area and the enhancement parameter is used as the corrected contrast corresponding to the small blood vessel area; and the fundus blood vessel image is enhanced based on the corrected contrast of all the small blood vessel areas.

8. The method for intelligent enhancement of fundus vascular images according to claim 1, characterized in that: The method for acquiring the blood vessel region comprises: Acquire a binary image of the fundus vascular image, extract a vascular skeleton of the binary image, and extract endpoints, branch points, and roots based on feature point detection; The blood vessel pixels between adjacent blood vessel branch points, between an end point and an adjacent blood vessel branch point, and between a root and an adjacent blood vessel branch point are recorded as one blood vessel region.

Images