TWI584226B - Method for extracting retina vessel image - Google Patents

Description

Retinal vascular image acquisition method

The present invention relates to a method of capturing an image; in particular, to a method of capturing a retinal blood vessel image.

In human or animal organs, the eye provides visual recognition, in other words, the color, shape, and even distance of the object can be identified through the eyes. A blind person may lose the judgment of space due to visual identification problems, and thus be at risk. Therefore, the eyes are called the window of the soul, which is to show its importance.

In the eye structure, the retina is a very important part. The blood vessels on the retina reveal a lot of important information, which is usually related to the judgment of eye diseases such as hypertension, diabetes, cardiovascular disease or glaucoma. Therefore, it is especially important to obtain retinal images and correctly interpret the blood vessels in the retina.

The original color retinal image captured in a conventional manner is shown in Figure 1. This color retinal image often has some problems that prevent the correct interpretation of the blood vessel image. For example, it is known that such color retinal images often have problems such as insignificant microvascular features, differences in image patches, rotational deformation, or size of blood vessel overlapping regions due to different viewing angles. Moreover, the contrast of such retinal images is low, so there is often extra noise in addition to the vascular image.

With the development of imaging technology, various methods for enhancing retinal vascular images have been proposed. However, these methods have the problem of not being able to accurately interpret the image, and greatly reduce the accuracy of retinal vascular imaging to judge various diseases.

For this reason, there is still an urgent need to develop a method for accurately taking out retinal blood vessel images in order to expand the application field of retinal blood vessel images.

In particular, the present invention provides a method of capturing retinal blood vessel images. By: extracting the G channel image of the color retinal image; converting the G channel image by the Gabor filter; binarizing the filtered image to determine the position of the blood vessel; corresponding to the position of each filtered image Each pixel is compared one by one, and the pixel with the largest value is taken out as a composite pixel; the synthesized retina image is formed by each synthesized pixel; the denoising step is performed to remove the noise of the non-vascular image to obtain the required Retinal vascular imaging; the retinal vascular image is processed by dilation pattern method to obtain enhanced retinal angiography, and the retinal vascular image can be accurately taken for subsequent interpretation.

In an embodiment, the present invention provides a method for capturing a retinal blood vessel image, comprising: reading a color retinal image; and separating the color retinal image into images corresponding to R, G, and B channels; Select the G channel image of the color retinal image; use a Gabor filter to convert the G channel image And replacing a plurality of filtered images, wherein each of the filtered images respectively corresponds to a different conversion angle; performing a region binarization step on each of the filtered images to extract a binarized image of each of the filtered images, and determining each of the two Whether the pixel points in the image are corresponding to the blood vessel part in the retinal image; in each filtered image, each pixel corresponding to the position is compared one by one, and the pixel with the largest value is taken as a composite pixel; Combining each synthetic pixel to obtain a synthetic retinal image; performing a denoising step to remove noise in the synthetic retinal image that is not part of the vascular site to obtain a retinal vascular image; and using a dilation type method for the retina The vascular image is processed to remove the enhanced retinal vascular image.

In the above method for capturing retinal blood vessel images, the denoising step is to remove noise that is not part of the blood vessel portion of the synthetic retinal image through a masking.

In the above method for capturing retinal blood vessel images, the Gabor filter system can be expressed by the following relationship: Equations (1) and (2) correspond to the real part and the imaginary part respectively; where: x' = x cos θ+ y sin θ; y' =- x sin θ+ y cos θ; λ: wavelength; θ: kernel function Direction; Ψ: phase offset; σ: Gaussian function standard deviation; γ: aspect ratio.

In the above method for capturing retinal blood vessel images, the G channel images are respectively converted into four angles (0°, 45°, 90°, 135°) using the above formulas (1) and (2); The real image and the imaginary image corresponding to each angle are added together to obtain respective filtered images corresponding to the respective angles, as shown in the following relation (3): among them, For real images, For the imaginary image, θ {0°, 45°, 90°, 135°}.

In the method for capturing retinal blood vessel images, the step of performing regional binarization on each filtered image includes: selecting an n*n (n=3, 5, 7, ...) template; obtaining an average value of all pixels in the template. And compare the values of the pixels in the template one by one; if the value of the pixel is larger than the average value, the value of the pixel is set to 255; if not, the value of the pixel is set to 0.

In the above method for capturing retinal blood vessel images, the step of processing the retinal blood vessel image by applying a dilation type method includes: setting a circular expansion of n*n (n=3, 5, 7, ...) Template; use a dilation pattern method to retinal vascular images according to a circular expansion template to remove enhanced retinal vascular images.

S101~S109‧‧‧Steps

101‧‧‧Filter image

102‧‧‧Filter image

103‧‧‧Filter image

104‧‧‧Filter image

S1‧‧‧ pixels

S2‧‧‧ pixels

S3‧‧‧ pixels

S4‧‧‧ pixels

1 is a view showing an original color retinal image; FIG. 2 is a flow chart showing a method for capturing retinal blood vessel images according to an embodiment of the present invention; and FIG. 3 is a view showing a color as shown in FIG. The G channel image of the retinal image; the 4th (a) to 4 (d) diagram shows the binarized image map corresponding to each conversion angle after the Ga channel filtering and the area binarization in the G channel image in FIG. 3; Figure 5 is a schematic diagram showing the extraction of synthesized pixels according to different conversion angles of Gabor filtering; Figure 6 is a diagram showing the combination of the synthesized pixels extracted in Figure 5 into a composite retinal image; and Figure 7 shows The final retinal angiogram obtained by the de-synthesis step of the synthetic retinal image in Fig. 6 and obtained by the expansion pattern method.

Hereinafter, a plurality of embodiments of the present invention will be described with reference to the drawings. For the sake of clarity, many practical details will be explained in the following description. However, it should be understood that these practical details are not intended to limit the invention. That is, in some embodiments of the invention, these practical details are not necessary. In addition, some of the conventional structures and elements are shown in the drawings in a simplified schematic manner in order to simplify the drawings.

Please refer to FIG. 2, which is a flow chart of a method for capturing retinal blood vessel images according to an embodiment of the present invention. The method of capturing retinal blood vessel images includes the following steps S101 to S109.

In step S101, a color retinal image is read.

In step S102, the color retinal image is separated into images corresponding to three channels of R, G, and B, respectively.

In step S103, only the G channel image of the color retinal image is selected.

Step S104: Convert a G channel image by using a Gabor filter to extract a plurality of filtered images, wherein each of the filtered image systems respectively corresponds to a different conversion angle.

Step S105: Perform a region binarization step on each filtered image to extract a binarized image of each filtered image, and determine whether a pixel in each binarized image corresponds to a blood vessel portion in the retinal image.

In step S106, in each of the filtered images, each pixel corresponding to the position is compared one by one, and the pixel having the largest value is taken out as a synthesized pixel.

Step S107, combining the synthesized pixels to obtain a synthetic retinal image.

Step S108, performing a denoising step to remove noise in the synthetic retinal image that does not belong to the blood vessel portion to obtain a retinal blood vessel image.

In step S109, the retinal blood vessel image is processed using a dilation type method to extract the enhanced retinal blood vessel image.

Each step will be described in detail later to facilitate understanding of the technical features of the present invention. At the same time, please refer to Figure 3 to Figure 7 together. Figure 3 shows the G channel image of the original color retinal image in Figure 1; the 4th (a) to 4 (d) shows the G channel image in Figure 3 after Gabor filtering and region binary After the transformation, the binarized retinal image map corresponding to each conversion angle is shown; the fifth figure shows the schematic diagram of extracting the synthesized pixel points according to the different conversion angles of the Gabor filter; and the sixth figure shows that the synthesized pixel points taken out in the 5 figure are merged into A synthetic retinal image; and Fig. 7 is a diagram showing the final retinal angiogram obtained by subjecting the synthetic retinal image of Fig. 6 to a denoising step and by an expansion pattern method.

In step S101, an original color retinal image is first read. As shown in FIG. 1, it can be seen that the shape is quite complicated, including blood vessels and other non-vascular parts.

Next, in step S102, the color retinal images are separated according to R, G, and B channels to form images corresponding to the R channel, the G channel, and the B channel.

In the step S103, only the G channel image is selected as the basis of the subsequent image processing, and the G channel image is as shown in FIG. The G channel image was selected because the eyeball was the clearest under the G channel.

Next, in step S104, the G channel image is converted using a Gabor filter. In this embodiment, the four different angles (0°, 45°, 90°, 135°) in the Gabor filter calculation program are converted, and four filtered images are taken out. The Gabor filter calculus program can be expressed in the following relation: Where equations (1) and (2) correspond to the real and imaginary parts, respectively; x' = x cos θ + y sin θ; y' = - x sin θ + y cos θ; λ: wavelength; θ: kernel function direction ;Ψ: phase offset; σ: Gaussian function standard deviation; γ: aspect ratio.

In the continuation, the G channel image is converted by the above formula (1) and formula (2); the real image corresponding to each angle and the imaginary image are added together to obtain the filtered images corresponding to the respective angles, as the following relationship Equation (3): among them, For real images, For the imaginary image, θ {0°, 45°, 90°, 135°}.

It should be noted that the selection of the number of angles mentioned above is not particularly limited, and an appropriate number of angles may be selected depending on the actual application.

After obtaining a plurality of filtered images corresponding to the different conversion angles, in step S105, a region binarization step is performed on each of the filtered images to extract a binarized image of each of the filtered images. The binarization method is applied to the so-called Otsu method, which first selects an n * n ( n = 3, 5, 7, ...) template; then obtains the average value of all the pixels in the template, and The values of the pixels in the template are compared one by one; if the value of the pixel is greater than the average value, the value of the pixel is set to 255; if not, the value of the pixel is set to zero. The binarized retinal images obtained in the above manner corresponding to the respective conversion angles are shown in Figures 4(a) to 4(d), and it can be seen that the figures are slightly different. By binarizing the image, it is possible to clearly determine whether the pixel in each binarized image corresponds to a blood vessel site in the original retinal image, so as to facilitate subsequent interpretation of the retinal blood vessel image.

Next, in step S106, for each of the filtered images formed in step S104, each pixel corresponding to the selected position is compared one by one, and the pixel having the largest value is taken as a value of the synthesized pixel. The manner in which the synthesized pixel points are taken out is shown in FIG. In Fig. 5, the filtered images 101 to 104 correspond to conversion angles of 0°, 45°, 90°, and 135°, respectively. The filtered images 101 to 104 are selected to correspond to the pixel points S1 to S4 at the same position. Compare the values of the pixel points S1~S4, and select the one with the largest value as the synthesized pixel point, and use its value as the value The value of the pixel corresponding to the same position for synthesizing the retina image. After all the pixel point positions are aligned one by one in this way, in step S107, all the synthesized pixel points can be combined into one synthetic retinal image, as shown in FIG.

Next, in step S108, a denoising step is performed to remove noise in the synthetic retinal image that does not belong to the blood vessel site to obtain a retinal blood vessel image. Based on the above-described step S105, the non-vascular site has been determined through the binarized image of each filtered image. Therefore, the denoising step is performed by a Masking method to remove noise from non-vascular sites in order to obtain a truly desired retinal blood vessel image.

Subsequently, in step S109, in order to further enhance the retinal vascular image features, the retinal vascular image is processed through a dilation type method to extract the enhanced retinal blood vessel image. The first step is to set a circular expansion template of n * n ( n = 3, 5, 7, ...); then use the dilation type method to retinal vascular image according to the circular expansion template, and finally take out Retinal vascular images as depicted in Figure 7. From Fig. 7, it can be seen that the image characteristics of the blood vessel site are more obvious after being treated by the dilation type method, which is favorable for the judgment of the retinal blood vessel image. The present invention is based on a transdermal morphology method in which retinal blood vessels can be connected.

In summary, in the method for capturing retinal blood vessel images disclosed by the present invention, a Gabor filter is used to extract filtered images corresponding to a plurality of conversion angles, and pixels having maximum values among the filtered images are taken out to form a synthetic retinal image. At the same time, the regional binarization step is performed to determine the position of the blood vessel, and the denoising step is performed and the retinal image is processed by the expansion pattern method, so that the retinal blood vessel image can be accurately taken out for subsequent interpretation.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and the present invention can be modified and modified without departing from the spirit and scope of the present invention. The scope is subject to the definition of the scope of the patent application attached.

S101~S109‧‧‧Steps

Claims (4)

A method for capturing retinal blood vessel images is performed in the following steps: reading a color retinal image; separating the color retinal image into images corresponding to R, G, and B channels; selecting only the color retinal image The G channel image is converted by using a Gabor filter to extract a plurality of filtered images, wherein each of the filtered images corresponds to a different conversion angle; and each of the filtered images is performed An area binarization step is performed to take out a binarized image of each of the filtered images, and determine whether a pixel point in each of the binarized images corresponds to a blood vessel portion in the retinal image; in each of the filtered images, Each pixel corresponding to the position is compared one by one, and the pixel with the largest value is taken as a composite pixel in each of the filtered images; combining the synthesized pixels to obtain a synthetic retinal image; performing a denoising step To remove noise from the vascular site of the synthetic retinal image to obtain a retinal vascular image; and to use a dilation type The morphological method processes the retinal vascular image to extract the enhanced retinal vascular image; wherein the step of binarizing the region for each of the filtered images comprises: selecting an n * n ( n =3, 5, 7 ,...) a template; and obtaining an average value of all the pixels in the template, and comparing the values of the pixels in the template one by one; if the value of the pixel is greater than the average, the pixel is The value is set to 255; if not, the value of the pixel is set to 0; wherein the step of processing the retinal blood vessel image by applying the dilation type method includes: setting an n * n ( n = a circular expansion template of 3, 5, 7, ...; and using the dilation pattern method to the retinal blood vessel image according to the circular expansion template to remove the enhanced retinal blood vessel image. The method for extracting retinal blood vessel images according to claim 1, wherein the performing the denoising step removes noise in the synthetic retinal image that does not belong to the blood vessel portion through a masking. The method for extracting retinal blood vessel images according to claim 1, wherein the Gabor filter is expressed by the following relationship: Equations (1) and (2) correspond to the real part and the imaginary part respectively; where: x' = x cos θ+ y sin θ; y' =- x sin θ+ y cos θ; λ: wavelength; θ: kernel function Direction; Ψ: phase offset; σ: Gaussian function standard deviation; γ: aspect ratio. The method for extracting retinal blood vessel images according to claim 3, wherein the G channel images are respectively subjected to four angles (0°) according to formula (1) and formula (2) of claim 3; , 45°, 90°, 135°) conversion; adding the absolute value of the real image corresponding to each angle to the imaginary image to obtain each of the filtered images corresponding to each angle, as shown in the following relation (3) : among them, (x, y) is the real image, (x, y) is the imaginary image, θ {0°, 45°, 90°, 135°}.