CN106651846B - Segmentation method of retinal blood vessel images - Google Patents

Abstract

The embodiment of the present invention provides a segmentation method of retinal blood vessel images. The method mainly includes: performing dual-scale matched filtering processing on retinal blood vessel images to obtain a fine-scale matched filtering response image and a coarse-scale matched filtering response image; segmenting the line support region from the fine-scale matched filtering response image, and using a local adaptive threshold method Binarize each line support area to segment thin blood vessel segments; apply a fixed ratio threshold algorithm to segment the coarse-scale matched filtered image to obtain thick blood vessel segments. The segmentation result of the thin blood vessel segment and the segmentation result of the thick blood vessel segment are fused to obtain a complete retinal blood vessel segmentation result, and the accuracy of the segmentation result is high.

Description

Segmentation method of retinal blood vessel images

technical field

The invention relates to the technical field of digital image processing, in particular to a segmentation method of retinal blood vessel images.

Background technique

Retinal blood vessels are the only blood vessels in the human body that can be directly observed non-invasively. Whether their shape, diameter, scale, branch angle change, and whether there is hyperplasia and exudation can reflect the lesions of systemic blood vessels. Diabetes, hypertension, cerebrovascular sclerosis and other diseases can lead to certain changes in retinal blood vessels. Therefore, the detection and analysis of retinal blood vessels has important clinical significance for the diagnosis and treatment of these diseases. However, due to the uneven gray distribution of retinal blood vessel images, the low contrast between the target blood vessels and the image background, and the contamination of image noise, the automatic segmentation of retinal blood vessels is very difficult.

At present, a retinal blood vessel image segmentation method in the prior art is: retinal blood vessel tracking method. Tolias et al. (1998) applied the fuzzy C-means method to establish seed points from the retinal disc, used the one-dimensional fuzzy model of the blood vessel section to track retinal blood vessels, and determined whether all the seed points belonged to blood vessels for final segmentation and extraction. This method can describe the retinal vascular network structure more comprehensively.

The disadvantages of the above retinal blood vessel tracking methods are: a large amount of computation, and inaccurate segmentation of branch points of blood vessels and blood vessels with low contrast. This method is sensitive to the selection of seed points, and the algorithm often terminates at the branch points of blood vessels, losing a large number of small retinal blood vessels.

Another method for segmenting retinal blood vessel images in the prior art is a classifier identification method. This method mainly preprocesses retinal blood vessels, establishes a feature space according to the different characteristics of blood vessels, selects a suitable classifier to train the samples, and then brings it into the test set to determine the final result. Staal (2004) adopted a ridge line detection method to perform supervised class recognition using eigenvectors in the neighborhood of the vessel centerline. Soares (2006) first extracted multi-scale 2D Gabor wavelet features of retinal blood vessel images, and then used a Bayesian classifier to identify retinal blood vessels.

The disadvantages of the above classifier identification method are: this classification method is sensitive to noise points, and the segmentation result has serious misclassification.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a method for segmenting a retinal blood vessel image, so as to effectively segment a thick blood vessel segment and a thin blood vessel segment from the retinal blood vessel image.

In order to achieve the above objects, the present invention adopts the following technical solutions.

A segmentation method of retinal blood vessel images, comprising:

Perform dual-scale matched filter processing on retinal blood vessel images to obtain fine-scale matched filter response images and coarse-scale matched filter response images;

Segment the line support area from the fine-scale matched filter response image, and use the local adaptive threshold method to binarize each line support area to segment the thin blood vessel segment;

A fixed-scale threshold algorithm is applied to segment the coarse-scale matched filtered image to obtain coarse blood vessel segments.

Further, performing dual-scale matched filtering processing on the retinal blood vessel image to obtain a fine-scale matched filtering response image and a coarse-scale matched filtering response image, including:

Extract the green channel of the three RGB channels of the color retinal blood vessel image;

A Gaussian function is used to simulate the cross-section grayscale curve of retinal blood vessels, and the following matched filter is obtained:

In the formula, K(x, y) is called the kernel function, σ is the deviation of the Gaussian function along the x-axis coordinate center, L is the length of the lightning channel where the Gaussian function is truncated along the y-axis, where x, y must satisfy | x|≤3σ, |y|≤L/2;

At 15° intervals, select 12 directions in the angle interval [0°, 180°] to create 12 matched filters;

Convolve the green channel in the color retinal blood vessel image with the 12 matched filters respectively to obtain a matched filter response image, and normalize and quantify the matched filter response image into a 256-level grayscale image , when the degree of deviation σ is less than the set threshold, the obtained grayscale image is used as the fine-scale matched filter response image; when the degree of deviation σ is not less than the set threshold, the obtained grayscale image is used as the coarse image Scale-matched filter response image.

Further, the described segmentation of the line support region from the fine-scale matched filter response image includes:

Calculate the gradient magnitude and gradient direction of each pixel in the fine-scale matched filtering image, sort all the pixels according to their gradient magnitudes, select the pixel with the highest gradient magnitude as the seed point, and use the gradient magnitude as the seed point. Pixels smaller than the set gradient threshold are excluded from the construction process of the online support area;

Using the region growing algorithm to generate several line support regions based on the seed points, each line support region includes a seed point and is a set of pixels with a gradient direction similar to the seed point, and each pixel point includes two states: using used and unused.

Further, the described use of the region growth algorithm based on the seed points to generate several line support regions, including:

Select an unused pixel point from the sorted list of pixel points as a seed point, take the gradient direction of the seed point as the initial angle θ _region of the line support region where the seed point is to be generated, and set the seed point The pixel points that are not used in the neighborhood of , and the error between the gradient direction and the region angle θ _region is between τ are added to the line support region, and the line support region is updated and calculated according to the updated pixel points. angle, where τ is the angle threshold;

The above processing process is repeatedly performed until no qualified pixel points are added to the line support area in the neighborhood of the seed point, and the minimum circumscribed rectangle corresponding to the line support area is extended.

Further, the described use of the local adaptive threshold method to perform binarization processing on each line support area to segment the thin blood vessel segments, including:

After dividing the fine-scale matched filter image into multiple line support regions, use the local adaptive threshold method to apply the Otsu algorithm to binarize each line support region, segment the foreground and background, and segment a single thin vessel segment;

The Otsu method searches for the optimal threshold to maximize the variance between the foreground and the background. Let t be the segmentation threshold of the foreground and the background, then calculate the probability w _0t and the average gray level u _0t of the foreground pixel, and the probability w _1t of the background pixel and the mean gray level is u _1t , the variance between foreground and background is expressed as:

g _t =w _0t ·(u _0t -u _t ) ² +w _1t ·(u _1t -u _t ) ² ,

Where u _t represents the total average gray level of the image, and the value range of t is 0-255. When the variance g _t is the largest, the difference between the foreground and the background is the largest, and the corresponding gray level t is the best threshold.

Further, the application of the fixed-scale threshold algorithm to segment the coarse-scale matched filtered image to obtain a thick blood vessel segment, including:

A fixed-scale threshold algorithm is applied to segment the coarse-scale matched filtered image to obtain a thick blood vessel image, and the threshold of the fixed-scale threshold algorithm is calculated by the following formula:

where r is the input parameter, representing the expected blood vessel ratio, Num is the frequency calculation function, and Total represents the total number of pixels;

When applying the fixed-scale threshold algorithm, first sort the pixels in the coarse-scale matched filtered image in descending order, search for the optimal threshold Tr, and perform two steps on the coarse-scale matched filtered image according to the optimal threshold Tr. Value processing, segment the foreground and background, segment out a single thick vessel segment.

Further, the method also includes:

A logical OR operation is applied to fuse the segmentation result of the thick blood vessel segment and the segmentation result of the thin blood vessel segment to obtain a complete segmentation result of the thick blood vessel segment and the thin blood vessel segment.

It can be seen from the technical solutions provided by the above embodiments of the present invention that the method of the embodiment of the present invention can effectively segment the thin blood vessels from the retinal blood vessel image by the ALT method, and can effectively segment the retinal blood vessel image by the FRT method. Complete thick blood vessels, fusion ALT method and FRT method can obtain complete retinal blood vessel segmentation results, and the segmentation results have high accuracy.

Additional aspects and advantages of the present invention will be set forth in part in the following description, which will be apparent from the following description, or may be learned by practice of the present invention.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a schematic diagram of an implementation principle of a method for segmenting a retinal blood vessel image based on a bi-scale nonlinear threshold of a blood vessel support area according to an embodiment of the present invention;

2 is a process flow diagram of a method for segmenting a retinal blood vessel image based on a bi-scale nonlinear threshold of a blood vessel support region provided by an embodiment of the present invention;

Fig. 3 is a rectangle expansion diagram provided by an embodiment of the present invention, wherein Fig. (a) represents the rectangle obtained by the region growing algorithm, Fig. Represents the expanded rectangle;

4 is a gray histogram of a line support area provided by an embodiment of the present invention;

Fig. 5 is a graph of fusion of thick and thin blood vessels provided by an embodiment of the present invention. Fig. (a) shows an example of the detection result of the local adaptive threshold, Fig. Fusion results.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but not to be construed as a limitation of the present invention.

It will be understood by those skilled in the art that the singular forms "a", "an", "the" and "the" as used herein can include the plural forms as well, unless expressly stated otherwise. It should be further understood that the word "comprising" used in the description of the present invention refers to the presence of stated features, integers, steps, operations, elements and/or components, but does not exclude the presence or addition of one or more other features, Integers, steps, operations, elements, components and/or groups thereof. It will be understood that when we refer to an element as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Furthermore, "connected" or "coupled" as used herein may include wirelessly connected or coupled. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as herein, are not to be taken in an idealized or overly formal sense. explain.

In order to facilitate the understanding of the embodiments of the present invention, the following will take several specific embodiments as examples for further explanation and description in conjunction with the accompanying drawings, and each embodiment does not constitute a limitation to the embodiments of the present invention.

The embodiment of the present invention proposes a segmentation method of retinal blood vessel images based on a bi-scale nonlinear threshold of the blood vessel support area. The method first uses two Gaussian filters of different scales to preprocess the retinal blood vessel images of the eye, and obtain two matched filter response images of different scales. Retinal vessel extraction is then performed on these two matched filtered response images at different scales, respectively. Finally, the extracted blood vessels are merged to obtain the final result. Specific steps are as follows:

A schematic diagram of the implementation principle of a method for segmenting retinal blood vessel images based on a bi-scale nonlinear threshold value of a blood vessel support area is proposed in an embodiment of the present invention, as shown in FIG. 1 , and a specific processing flow is shown in FIG. 2 , including the following processing steps:

Step S210, performing double-scale matching filtering (double-scale filtering, DSF) processing on the retinal blood vessel image to obtain a fine-scale matched filtering response (fine matched filtering response, FMFR) image and a coarse matched filtering response (coarse matched filtering response, CMFR) image.

Different scales of Gaussian matched filters have different enhancement effects on thin blood vessels and thick blood vessels. More specifically, the small-scale filter is beneficial to highlight the thin blood vessels and extract the skeleton of the thick blood vessels; while the coarse-scale filter is beneficial to enhance the thick blood vessels and blur the peripheral thin blood vessels. Therefore, we designed two types of Gaussian kernel functions, fine-scale and coarse-scale, and applied them to the green channel of retinal blood vessel images, respectively, to obtain a fine-scale matched filter response image and a coarse-scale matched filter response image.

Among the three RGB (Red, Green, Blue) channels of the color retinal blood vessel image, the green channel component image has the highest contrast between blood vessels and the background, which is more conducive to blood vessel segmentation. Therefore, the present invention first extracts the green channel of the color retinal blood vessel image.

In the retinal blood vessel image, the brightness of the pixel in the center of the blood vessel is relatively small, and the brightness of the pixels on both sides is relatively high. Therefore, Gaussian matched filtering method is often used to improve image contrast. Assuming that the retinal blood vessels are straight line segments of equal width, the length is L and the width is 2σ, we use a Gaussian function to simulate the grayscale curve of the cross-section of the retinal blood vessels, so as to obtain the following matched filter:

In the formula, K(x, y) is called the kernel function, σ is the deviation of the Gaussian function along the x-axis coordinate center, and L is the length of the lightning channel where the Gaussian function is truncated along the y-axis. In order to make the matching more accurate, where x, y must satisfy |x|≤3σ, |y|≤L/2. According to the experimental results, we set L=7.

Because the vessel direction is arbitrary, we create 12 matched filters by considering 12 directions in the angular interval [0°, 180°] at 15° intervals.

The retinal blood vessel image is convolved with these 12 Gaussian kernels respectively, and the value of each pixel in the matched filter response image is equal to the maximum convolution value. To facilitate subsequent processing, the matched filter response images are normalized and quantized into 256-level grayscale images.

We can observe that the retinal blood vessel images contain both the thick blood vessels in the optic disc attachment and the thin blood vessels in the periphery. When applying matched filtering, if you choose a relatively small value range of σ, this relatively small range is the width of the blood deviation tube mentioned above or the deviation of the Gaussian function along the x-axis coordinate center, which is about 1.3 to 1.6 pixel, the thin blood vessels in the filtering result image are more likely to be strengthened, and the thick blood vessels are corroded; on the contrary, if you choose a relatively large value range of σ, the relatively large value range is 2.0 ~ 2.4 pixels, then the thick blood vessels are strengthened, and the thin blood vessels are eroded. Blood vessels are blurred.

Therefore, the embodiment of the present invention proposes a dual-scale matched filtering method. The fine-scale matched filter selects a smaller σ to enhance thin blood vessels while suppressing noise and smoothing background regions. The response results produced by the fine-scale matched filter are quantized to obtain the FMFR image, which will be the input to the adaptive threshold segmentation algorithm. On the contrary, the coarse-scale matched filter selects a larger σ to enhance the coarse vessel part, resulting in a CMFR image. Because the edge part of the thick blood vessel is easily corroded in the fine-scale matched filtered image, but the thick blood vessel is easier to be completely segmented from the coarse-scale matched filtered image, so the CMFR image will be used for the fusion of segmentation results.

Step S220: Segment the Vessel Support Region (VSR) from the fine-scale matched filter response image, use the adaptive local thresholding (ALT) method and apply the Otsu algorithm to binarize each VSR, and segment out a single VSR. segment of thin blood vessels.

After matched filtering, the contrast of FMFR images is enhanced, especially the stains and lesion areas are suppressed. However, the gray distribution of blood vessels in FMFR images is still relatively scattered, and the gray values of some blood vessels overlap with the background gray values. In theory, we cannot find a global threshold to linearly segment vessels and background. VSR refers to a rectangular area containing a vessel segment, which can be automatically detected by an algorithm. In a local VSR whose histogram has a distinct bimodal nature, we can apply automatic thresholding algorithms (such as Otsu) to segment vessels and background. The process first automatically detects all VSR regions in FMFR, and then applies the Otsu algorithm to segment each VSR region, while setting the pixels of all non-VSR regions as the background, and finally obtains a fine vessel segmentation (FVS).

S2-1: Line support area detection

Since retinal blood vessels are sensitive to the small scale parameter σ, the present invention first calculates the matched filtered image under the condition of scale σ=1.3, and then performs the following processing on this basis.

S2-1-1: Line Support Area Generation

First, the gradient magnitude and gradient direction of each pixel in the fine-scale matched filtering image are calculated, and then all pixels are sorted according to their gradient magnitudes. Strong edge points or regions generally have relatively high gradient amplitudes, and usually pixels at the edge of retinal blood vessels have the highest gradient amplitudes, so the pixels with the highest gradient amplitudes are first selected as seed points. In the calculation process, the pixels whose gradient amplitude is less than q (0.4 is selected in the present invention) will be rejected to participate in the construction process of the line support area.

The present invention uses the region growing algorithm to generate several line support regions, and each line support region is a set of pixels with a similar gradient direction to the seed point. Each pixel includes two states, used and unused. The initial state sets all pixels to unused.

The region growing algorithm first selects an unused pixel from the sorted list as a seed point, and the unused pixel in the neighborhood of this pixel and the error between its gradient direction and the region angle θ _region is between τ will be added to it. in this area. The range of the experimental τ in this paper is about 18° to 24°, and it is generally set to 22 by default.

The initial angle θ _region of the region is the gradient direction of the seed point. Each time a new pixel is added to the region, the region's angle is updated. The angle of the area is then updated to:

θ _j represents the vertical direction of the gradient of pixel j.

i(x, y) represents the gray value at the pixel (x, y) point, g _x (x, y), g _y (x, y) represent the gradient of the pixel (x, y) in the x and y directions, respectively value.

And so on until there are no more pixels to add to the rectangle.

The line support area obtained earlier is represented by a minimum enclosing rectangle. Thus, some basic information of the rectangle can be obtained. FIG. 3 is a rectangle expansion diagram provided by an embodiment of the present invention, including the coordinates of the center point of the rectangle, the length and width of the rectangle, and the main direction. Figure (b) represents the expansion of the rectangle in the abscissa and ordinate directions, and Figure (c) represents the expanded rectangle.

The expansion of the rectangle is divided into two steps. The first step is to first perform equal expansion on both the horizontal and vertical directions of the rectangle, and the expansion range is selected as half of the width of the rectangle). In this way, every time a rectangle is expanded, the state of the pixels in the rectangle is set to Used, and the points that are set to Used next time will not be selected as seed points. Step 2: On the basis of the previously expanded rectangle, perform equal expansion both horizontally and vertically, and the expansion range is the same as the previous fixed value. Thresholding on the expanded rectangle will solve the problem of no intersection between the two rectangles, since those pixels added after the second expansion can be selected as seed points.

Algorithm 1. Region Growing Algorithm

S2-1-2: Rectangular Growth

The line support area obtained by the regional growth algorithm can well cover most of the lightning channel area, but it has two shortcomings: (1) The gradient values on both sides of the lightning channel are relatively large, so the seed point of the regional growth is generally the lightning channel. The pixel points of the two edges cause two rectangles to appear at the same cross-section of the lightning channel; (2) the seed points for the growth of the next area are not necessarily selected from the generated rectangular area, so the newly generated area follows The previous area does not necessarily have an overlapping area. That is, the rectangle may not be continuous along the direction of the lightning channel, so that all the foreground regions cannot be fully included.

Therefore, the present invention proposes a rectangle expansion algorithm, that is, two-way expansion is performed on the basis of the original rectangle, and the expansion range is selected as half of the width of the rectangle. Among them, vertical expansion (along the direction of blood vessel development) can solve the problem of discontinuity of the rectangular area, and lateral expansion (perpendicular to the development direction of blood vessels) can achieve the effect of merging two rectangles.

Figure 3 shows a schematic diagram of the expansion of the rectangle, in which point O is the center point of the rectangle, theta is the main direction of the rectangle, Length and Width are the length and width of the rectangle, respectively, and P and Q are the center point on the width of the rectangle. The sizes of x1 and x2, and y1 and y2 are divided into 9 cases.

S2-2: Line support region threshold segmentation

Applying the above VSR detection algorithm, one MFR image can be segmented into multiple VSRs.

In step S230, the ALT method applies the Otsu algorithm to perform binarization processing on each VSR to segment the foreground and background, and segment a single thin vessel segment. Intuitively, each VSR region is an image block with very obvious contrast, and the blood vessel segment and the background have obvious differences in gray space. From a statistical point of view, the distribution of intensity values in the VSR region has bimodality, that is, background pixels and blood vessel pixels are concentrated in their respective intensity intervals. Figure 4 shows the grayscale histograms of two randomly selected VSRs. Our more experimental results all show that the intensity distribution of VSR is bimodal.

Otsu is a classic automatic thresholding method, which has a good segmentation effect on images with bimodal distribution. The principle of Otsu's method is to search for the optimal threshold to maximize the variance between foreground and background. Assuming t is the segmentation threshold of foreground and background, the probability w _0t and average gray level u _0t of foreground pixels can be calculated, and the probability w _1t and average gray level of background pixels are u _1t . The variance between foreground and background can be expressed as:

g _t =w _0t ·(u _0t -u _t ) ² +w _1t ·(u _1t -u _t ) ² ,

Where u _t represents the total average gray level of the image, and the value range of t is 0-255. When the variance g _t is the largest, the difference between the foreground and the background is the largest, and the corresponding gray t is the best threshold.

Step S240 , segment the CMFR image by applying a fixed ratio threshold algorithm to obtain a segmentation map of thick blood vessels, and fuse the segmentation methods of thick and thin blood vessels to obtain a complete segmentation result of thin blood vessels and thick blood vessels.

The fusion consists of two main steps: First, a fixed-scale threshold algorithm is applied to segment the CMFR image to obtain a coarse vessel segmentation (CVS). Then, FVS and CVS are fused by logical OR operation, so that the fusion result contains both thin vessels and complete thick vessels.

ALT can detect thin blood vessels, but the thick blood vessels segmented by ALT often only contain their skeleton, while omitting the pixels of their peripheral parts. Figure 5(a) shows an example of the detection results of ALT. In order to improve the detection performance of ALT, the present invention proposes a fusion method of thin blood vessels and thick blood vessels. The method firstly applies the fixed-ratio thresholding algorithm (FRT) to segment the coarse-scale matched filtered image to obtain the coarse blood vessel image, and then fuses the thin blood vessel image and the thick blood vessel image to obtain the final thick and thin blood vessel fusion result.

The fixed-scale threshold algorithm is a simple prior-based binarization method. Intuitively, the retinal image has obvious structure, that is, it consists of linear blood vessels and a flat background, and the proportion of blood vessels is often low. From a statistical point of view, the mean and variance of vessel pixel proportions in DRIVE were 8.43% and 1.38%, respectively, and the mean and variance of vessel pixel proportions in STRES were 7.6% and 3.15%, respectively. The two datasets, DRIVE and STARES, only use the average value as the evaluation index, and the variance is removed. The average value of DRIVE is 12.7%, and the average value of STRES is 10.4%.

It should be pointed out that there are some retinal images of lesions in STARES, so the variance of the proportion of blood vessels is relatively large. In the coarse-scale matched filtered image, the response value of coarse vessel pixels is larger than the response values of thin vessel and background pixels. Therefore, the threshold of the FRT method can be calculated by the following formula:

where r is an input parameter representing the expected vessel proportion. Num is the frequency calculation function, and Total represents the total number of pixels. When implementing the algorithm, the bucket sorting algorithm is used to sort the pixels in the coarse-scale matched filtered image in descending order, then the optimal threshold Tr is searched, and finally the coarse-scale matched filtered image is binarized according to Tr. Figure 5(b) shows the segmentation result of FRT. It can be observed that this result segmented thick vessels relatively completely, although it lost the details of thin vessels.

In conclusion, ALT is good at segmenting thin blood vessels, while FRT can segment complete thick blood vessels. Therefore, the results of fusing the two methods can expect relatively perfect segmentation performance. Simply, the present invention applies a logical OR operation to fuse the results of ALT and FRT. That is, one of the gray values of the corresponding pixel points in the two images is 255, then the gray value of the corresponding pixel points in the resulting image is 255, and only when the gray values of the corresponding position pixels in the two images are 0, the gray value of the pixel at the corresponding position in the resulting image is 0.

In addition, in order to eliminate the interference of background noise and some diseased tissues, the area with an area of less than 10 pixels was removed. Figure 5(c) shows the final fusion result.

To sum up, the method of the embodiment of the present invention can effectively segment the thin blood vessels from the retinal blood vessel image by the ALT method, and can effectively segment the complete thick blood vessel from the retinal blood vessel image by the FRT method, and fuse the ALT method and the FRT method. The method can obtain complete segmentation results of retinal blood vessels, and the segmentation results have high accuracy.

At present, most methods at home and abroad only extract blood vessels for normal and well-imaged retinal images. For retinal images with low contrast or lesions, because the pixel gray values of the blood vessels and the background area are relatively close, the existing traditional Most of the methods fail to correctly segment blood vessels from the background. On the other hand, the present invention divides the blood vessels into thick and thin blood vessels by using Gaussian matching filtering of large and small scales, and performs enhancement processing respectively, and the effect is obvious. For thin blood vessels, the gradient size and direction of pixels are used, and the method based on region growth and rectangle expansion has good local adaptability and can quickly determine the foreground region. For thick vessels, using a global fixed threshold can better preserve the trunk portion.

Those of ordinary skill in the art can understand that the accompanying drawing is only a schematic diagram of an embodiment, and the modules or processes in the accompanying drawing are not necessarily necessary to implement the present invention.

From the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solutions of the present invention can be embodied in the form of software products in essence or the parts that make contributions to the prior art. The computer software products can be stored in storage media, such as ROM/RAM, magnetic disks, etc. , CD, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the apparatus or system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference may be made to some descriptions of the method embodiments for related parts. The apparatus and system embodiments described above are only illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, It can be located in one place, or it can be distributed over multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited to this. Substitutions should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (4)

1. a segmentation method of retinal blood vessel image, is characterized in that, comprises: Perform dual-scale matched filtering processing on the retinal blood vessel image to obtain a fine-scale matched filtering response image and a coarse-scale matched filtering response image, including: when applying matched filtering, select a relatively small value range σ, this relatively small range is 1.3～ If the value is 1.6 pixels, the thin blood vessels in the filtering result image are more likely to be strengthened, and the thick blood vessels are corroded; on the contrary, choose a relatively large value range of σ, which is 2.0 to 2.4 pixels, then the thick blood vessels are strengthened, The thin blood vessels are blurred, where σ is the width of the deviation of the blood vessel or the deviation of the Gaussian function along the x-axis coordinate center; Segment the line support area from the fine-scale matched filter response image, use the local adaptive threshold method to binarize each line support area, and segment the thin blood vessel segments, including: Calculate the gradient magnitude and gradient direction of each pixel in the fine-scale matched filtering image, sort all the pixels according to their gradient magnitudes, select the pixel with the highest gradient magnitude as the seed point, and use the gradient magnitude as the seed point. The pixel points smaller than the set gradient threshold are excluded from the construction process of the online support region; several line support regions are generated based on the seed point using the region growing algorithm, and each line support region includes a seed point, and is one and the seed point. A collection of pixels with similar gradient directions, each pixel includes two states: used and unused; After dividing the fine-scale matched filter image into multiple line support regions, use the local adaptive threshold method to apply the Otsu algorithm to binarize each line support region, segment the foreground and background, and segment a single thin vessel segment; The Otsu method searches for the optimal threshold to maximize the variance between the foreground and the background. Let t be the segmentation threshold of the foreground and the background, then calculate the probability w _0t and the average gray level u _0t of the foreground pixel, and the probability w _1t of the background pixel and the mean gray level is u _1t , the variance between foreground and background is expressed as: g _t =w _ot ·(u _0t -u _t ) ² +w _1t ·(u _1t -u _t ) ² Where u _t represents the total average gray level of the image, and the value range of t is 0-255. When the variance g _t is the largest, the difference between the foreground and the background is the largest, and the corresponding gray level t is the best threshold; Applying a fixed-scale threshold algorithm to segment the coarse-scale matched filtered image to obtain a thick blood vessel segment, including: applying a fixed-scale threshold algorithm to segment the coarse-scale matched filtered image to obtain a thick blood vessel image, the fixed-scale threshold algorithm The threshold of is calculated by the following formula: where r is the input parameter, representing the expected blood vessel ratio, Num is the frequency calculation function, and Total represents the total number of pixels; When applying the fixed-scale threshold algorithm, first sort the pixels in the coarse-scale matched filtered image in descending order, search for the optimal threshold Tr, and perform two steps on the coarse-scale matched filtered image according to the optimal threshold Tr. Value processing, segment the foreground and background, segment out a single thick vessel segment. 2. method according to claim 1, is characterized in that, described to retina blood vessel image is carried out double-scale matched filter processing, obtains fine-scale matched filter response image and coarse-scale matched filter response image, comprises: Extract the green channel of the three RGB channels of the color retinal blood vessel image; A Gaussian function is used to simulate the cross-section grayscale curve of retinal blood vessels, and the following matched filter is obtained: In the formula, K(x, y) is called the kernel function, σ is the deviation of the Gaussian function along the x-axis coordinate center, L is the length of the lightning channel where the Gaussian function is truncated along the y-axis, where x, y must satisfy | x|≤3σ,|y|≤L/2; At 15° intervals, select 12 directions in the angle interval [0°, 180°] to create 12 matched filters; Convolve the green channel in the color retinal blood vessel image with the 12 matched filters respectively to obtain a matched filter response image, and normalize and quantify the matched filter response image into a 256-level grayscale image , when the degree of deviation σ is less than the set threshold, the obtained grayscale image is used as the fine-scale matched filter response image; when the degree of deviation σ is not less than the set threshold, the obtained grayscale image is used as the coarse image Scale-matched filter response image. 3. The method according to claim 2, wherein the generating several line support regions based on the seed point using a region growing algorithm, comprising: Select an unused pixel point from the sorted list of pixel points as a seed point, take the gradient direction of the seed point as the initial angle θ _region of the line support region where the seed point is to be generated, and set the seed point The pixel points that are not used in the neighborhood of , and the error between the gradient direction and the region angle θ _region is between τ are added to the line support region, and the line support region is updated and calculated according to the updated pixel points. angle, where τ is the angle threshold; The above processing process is repeatedly performed until no qualified pixel points are added to the line support area in the neighborhood of the seed point, and the minimum circumscribed rectangle corresponding to the line support area is extended. 4. The method according to any one of claims 1 to 3, wherein the method further comprises: A logical OR operation is applied to fuse the segmentation result of the thick blood vessel segment and the segmentation result of the thin blood vessel segment to obtain a complete segmentation result of the thick blood vessel segment and the thin blood vessel segment.