JP2018102586A - Fundus image processing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a fundus image processing device capable of obtaining an image by which determination of an artery-vein diameter ratio is facilitated.SOLUTION: A fundus image processing device comprises: blood vessel region enhancement means 10 for acquiring a blood vessel enhancement image in which as a probability of being a blood vessel region is higher, a higher value is set, by enhancing a blood vessel region to a color fundus image; hue conversion means (20, 30, 40) for acquiring a hue conversion image in which an RGB balue is converted into a hue value of an L gradation by converting the RGB value acquired by referring to a corresponding pixel of the color fundus image to each pixel in the blood vessel region into the hue value of the L gradation; threshold calculation means 50 for calculating a threshold of binarization by using a histogram calculated on the basis of the pixel corresponding to the blood vessel region of the hue conversion image; and artery-vein identification image generation means 60 for, on the basis of a comparison result between the hue value of the L gradation of each pixel corresponding to the blood vessel region of the hue conversion image and the threshold, acquiring an artery-vein identification image to which an attribute of an artery or a vein is applied.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a fundus image processing technique, and more particularly to a technique for supporting a process for obtaining an arteriovenous diameter ratio using a fundus image.

  In so-called metabolic syndrome such as hypertension, dyslipidemia, diabetes, and arteriosclerosis that are typical lifestyle-related diseases, blood pressure, lipid, and blood glucose can be easily measured and self-examination is possible. However, at present, arteriosclerosis measurement (commonly known as blood vessel age test) is specialized in circulatory organs equipped with testing equipment such as PWV test (pulse wave velocity), ABI test (ankle brachial blood pressure ratio), and carotid echocardiography. Measurement is not possible if it is not a clinic.

  On the other hand, a technique for simply measuring hypertension and arteriosclerosis by measuring the fundus blood vessel diameter using a fundus camera owned by an ophthalmology clinic is also known. The fundus is the only place in the human body where blood vessels can be directly observed. If it is assumed that hardening of the fundus artery proceeds in the same manner as the arteries throughout the body, it is possible to measure arteriosclerosis by taking fundus photographs. Fundus cameras are becoming smaller and less expensive, and overseas, special lenses that can be photographed with smartphones are already being sold, and infrastructure for self-photographing of fundus photographs is being prepared.

  Scheie classification has been proposed as a method for measuring hypertension and arteriosclerosis using fundus photographs. The main measurement items are the arteriovenous diameter ratio, the arteriovenous intersection ratio, and the arterial reflex ratio. A measurement support method corresponding to these has also been proposed (see Patent Document 1). In addition, a method of fully automatically measuring the arteriovenous diameter ratio has been proposed (see Patent Document 2).

JP 2007-319403 A Japanese Patent Laid-Open No. 10-243924

  In the technique described in Patent Document 1, regarding the measurement of the blood vessel width, a point where the difference in pixel value from the adjacent pixel is large is regarded as a boundary from the center instructed in a circle by the user, and the distance to the nearest boundary point is calculated. Based on this, the radius of the blood vessel is calculated. However, when the blood vessel boundary is blurred due to the condition of the fundus, there is a problem that the boundary point cannot be specified and the blood vessel width is measured to be small. In the technique described in Patent Document 2, there is a difference in brightness between an artery and a vein (both brightness is lower than the background, but the vein is significantly lower than the artery), and binarization is performed with the same threshold value to extract blood vessels. Then, the artery width is small and the vein width is large. Therefore, the arteriovenous diameter ratio cannot be measured correctly. Also, a method of automatically performing arteriovenous discrimination has been proposed and a function of automatically correcting an erroneous determination location based on the continuity of a blood vessel has been proposed.

Due to the problems as described above, at present, a method in which an ophthalmologist applies a gauge on a fundus photograph and visually measures it is common. Therefore, the arteriovenous diameter ratio is a rough value such as 1/2 or 1/3, and there is a problem that the degree of progress cannot be compared quantitatively. On the other hand, even if the reading accuracy is increased, the variation due to the subjectivity of the measurer and the variation due to the measurement location increase, and it is necessary to sample a large number of measurement locations, but the workload of the ophthalmologist is large and not realistic.

  Therefore, an object of the present invention is to provide a fundus image processing apparatus capable of obtaining an image in which an artery and a vein are identified in order to facilitate measurement of an arteriovenous diameter ratio and the like.

In order to solve the above problems, in the first aspect of the present invention,
A blood vessel region enhancement means for obtaining a grayscale blood vessel enhancement image in which a higher value is set for a pixel having a higher probability of being a blood vessel region with respect to a color fundus image;
A pixel having a pixel value greater than or equal to a predetermined value in the blood vessel enhanced image is extracted as a pixel of the blood vessel region, and an RGB value is obtained by referring to a corresponding pixel of the color fundus image for each pixel of the blood vessel region. Hue conversion means for converting the acquired RGB value into a hue value of L gradation (L is a positive integer) and obtaining a hue conversion image in which the pixel value is converted to the hue value of L gradation;
Threshold calculation that calculates a histogram of L gradation based on pixels corresponding to the blood vessel region of the hue conversion image, and calculates a threshold value for binarization by a discriminant analysis method using the calculated histogram Means,
An arteriovenous identification image to which an attribute including at least an artery or a vein is given is obtained based on a comparison result between an L tone hue value of each pixel corresponding to the blood vessel region of the hue conversion image and the threshold value. An arteriovenous identification image generating means;
There is provided a fundus image processing apparatus characterized by comprising:

In the second aspect of the present invention,
The hue conversion means includes
A pixel having a pixel value greater than or equal to a predetermined value in the blood vessel enhanced image is extracted as a pixel of the blood vessel region, and an RGB value is obtained by referring to a corresponding pixel of the color fundus image for each pixel of the blood vessel region. The obtained RGB values are converted into hue values of K (K is an integer greater than L) gradation, and a multi-gradation hue conversion image in which pixel values are converted to hue values of K gradation is obtained. Gradation hue conversion means;
A central gradation area defining means for defining a central gradation area within a range of cmin (cmin is a positive integer) to cmax (cmax is larger than cmin and smaller than K-1) within the range of the K gradation;
With respect to the multi-tone hue conversion image, the hue conversion image of L gradation having a value from 0 to L-1 from 0 to L-1 is obtained by saturating cmin or less to 0 and cmax or more to L-1. A gradation reduction means for conversion;
It further has these.

  In the third aspect of the present invention, the multi-gradation hue conversion unit converts the acquired RGB values into hue values for all pixels in the blood vessel region, and then sets the minimum hue value to 0. The maximum hue value is converted into an integer hue value of K gradation so as to be converted into K-1.

  Further, in the fourth aspect of the present invention, the central gradation area defining means includes the blood vessel in the high-accuracy histogram calculated based on pixels corresponding to the blood vessel area of the multi-gradation hue conversion image. The total number of pixels corresponding to the region is defined as Ct, the median value is hc, and a range including pixels of Ct / 4 on both sides of the median value hc is defined as the median gradation region.

In the fifth aspect of the present invention,
The blood vessel region enhancement means includes
Grayscale conversion means for obtaining a grayscale fundus image based on at least one of the color components of the color fundus image;
For each pixel of the grayscale fundus image, an average value of pixel values of neighboring pixels included in a predetermined range is calculated, and a value obtained by subtracting the average value from the predetermined value is added to the pixel value of the pixel. Image flattening means for creating a flattened image,
Linear component enhancement means for performing an opening process on the flattened image using a predetermined structural element and creating a linear component enhanced image in which the luminance of the blood vessel candidate region is equal to or higher than a certain level;
In the linear component emphasized image, a pixel value that does not satisfy the predetermined condition is replaced with the minimum gradation value of the converted blood vessel emphasized image, and the pixel value that satisfies the predetermined condition is the minimum gradation value. Pixel gradation conversion means for correcting the gradation of the pixel so as to be in the range of the maximum value from the value near the value, and creating a blood vessel emphasized image;
It is characterized by having.

  In the sixth aspect of the present invention, the hue conversion unit extracts pixels having a pixel value of 128 or more as pixels of a blood vessel region when the blood vessel emphasized image has a pixel value of 256 gradations. And

  In the seventh aspect of the present invention, the multi-gradation hue conversion means converts the acquired RGB values into the K-gradation hue values by using the values corresponding to R, G, and B as r, g , B, α as a predetermined real value, h = (g−b) α / (2r−g−b), and 2r−g−b = 0, h = 0 and 2r−g−b If <0, h = 0 for normal eyes and h = (g−b) α / (2r−g−b) for cataract eyes based on the obtained value of h. Features.

  In the eighth aspect of the present invention, the threshold value calculation means sets the L gradation to 256 gradations, and uses the range of gradation 1 to gradation 254 of the calculated histogram to perform the process of discriminant analysis method. And the threshold value is calculated.

  Also, in the ninth aspect of the present invention, the arteriovenous identification image generating means sets the L gradation to 256 gradations and the threshold value to Sav, and each corresponding to the blood vessel region of the hue conversion image. The pixel value Hue is converted to a pixel value in the range of −255 to 255 by performing an operation of (Hue−Sav) × 2 for the hue value Hue of the pixel, and a pixel value 0 is set for each pixel not corresponding to the blood vessel region. If the pixel value is positive, the pixel is given the artery attribute, if the pixel value is negative, the pixel is given the vein attribute, and if the pixel value is 0, the pixel is not included in the blood vessel region Then, the arteriovenous identification image is generated.

In the tenth aspect of the present invention,
With respect to the arteriovenous identification image generated by the arteriovenous identification image generating means, sequentially search for pixels having pixel values from (maximum pixel value-1) to 1 as correction target pixels,
The absolute value of the pixel value of each of the correction target pixels searched for is ZT, the maximum absolute value of the pixel values among the eight neighboring pixels of the correction target pixel is ZM, and the arterial attribute is selected among the eight neighboring pixels. ZM> ZT, where Cp is the sum of absolute values of the pixel values of pixels having Cp and Cn is the sum of absolute values of the pixel values of pixels having the vein attribute among the eight neighboring pixels,
When the attribute of the pixel having the maximum value ZM satisfies Cp> Cn in the artery, or the attribute of the pixel having the maximum value ZM satisfies Cp <Cn in the vein, the attribute of the correction target pixel is set to the pixel having the maximum value ZM. An arteriovenous identification image correcting unit that sequentially corrects all of the searched correction target pixels so as to replace the attribute and replace the absolute value of the pixel value of the correction target pixel with ZM. It is characterized by having.

In the eleventh aspect of the present invention,
A pixel block of (2N + 1) × (2N + 1) pixels (N ≧ 1) is sequentially extracted from the arteriovenous identification image generated by the arteriovenous identification image generating means, and has an arterial attribute in the pixel block. If the total number of pixels is Cpa, and the total number of pixels having the vein attribute in the pixel block is Cna,
When Cpa> Cna, the attribute of all pixels in the pixel block is set as an artery, and when Cpa <Cna, the correction is performed so that the attributes of all pixels in the pixel block are set as veins. It further has a vein identification image correcting means.

  According to a twelfth aspect of the present invention, there is provided a program for causing a computer to function as the fundus image processing apparatus.

  According to the present invention, it is possible to obtain an image for facilitating determination of an arteriovenous diameter ratio.

1 is a hardware configuration diagram of a fundus image processing apparatus 100 according to an embodiment of the present invention. 1 is a functional block diagram illustrating a configuration of a fundus image processing apparatus according to an embodiment of the present invention. It is a flowchart which shows the process outline | summary of the fundus image processing apparatus according to an embodiment of the present invention. It is a figure which shows an example of a full-color fundus image and a blood vessel emphasis image. It is a figure which shows a 2048 gradation hue histogram when K = 2048. It is a figure which shows the center gradation area | region extracted based on the histogram of FIG. It is a figure which shows an example of L gradation conversion image Hue (x, y) when L = 256. It is a figure which shows a 256 tone hue histogram when L = 256. It is a figure which shows an example of correction | amendment of the arteriovenous identification image Hues (x, y) by a 1st method. It is a figure which shows an example of correction | amendment of the arteriovenous identification image Hues (x, y) by a 2nd method. It is a figure which shows the example of a display of the arteriovenous determination result image AVsep. It is a figure which shows the example of a display of the arteriovenous determination result image AVsep after correction | amendment by a 1st method. It is a figure which shows the example of a display of the arteriovenous determination result image AVsep after correction | amendment by the 1st method and the 2nd method. It is a figure which shows another example of the full-color fundus image of a cataract eye corresponding to FIG. 4, and a blood vessel emphasized image. It is a figure which shows the example of a display of the arteriovenous determination result image AVsep with respect to the case shown in FIG. FIG. 16 is a diagram illustrating a display example of an arteriovenous determination result image AVsep after correction by the first method and the second method for the case illustrated in FIG. 15. 3 is a functional block diagram showing details of a blood vessel region emphasizing means 10. FIG. It is a flowchart which shows the detail of the blood vessel region enhancement process of step S100 by the blood vessel region enhancement means 10. It is a figure which shows the relationship between the pixel (x, y) on a gray scale fundus image, and the vicinity pixel for calculating average value Mean (x, y). It is a flowchart which shows the detail of the linear component emphasis process of step S130. It is a figure which shows an example of the linear structural element used by an opening process. It is a figure which shows an example of the circular structural element used by an opening process. It is a figure which shows the specific pixel value of the linear structure element in the case of direction d = 1. It is a figure which shows the specific pixel value of the linear structure element in the direction of d = 2.

DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the drawings.
<1. Device configuration>
FIG. 1 is a hardware configuration diagram of a fundus image processing apparatus according to an embodiment of the present invention. The fundus image processing apparatus 100 according to the present embodiment can be realized by a general-purpose computer. As shown in FIG. 1, a CPU (Central Processing Unit) 1 and a RAM (Random Access Memory) that is a main memory of the computer. 2, a large-capacity storage device 3 such as a hard disk or flash memory for storing programs and data executed by the CPU 1, an instruction input I / F (interface) 4 such as a keyboard and a mouse, and a data storage medium A data input / output I / F (interface) 5 for data communication with an external apparatus and a display unit 6 which is a display device such as a liquid crystal display are provided, and are connected to each other via a bus.

  FIG. 2 is a functional block diagram illustrating a configuration of the fundus image processing apparatus according to the present embodiment. In FIG. 2, 10 is a blood vessel region emphasizing unit, 20 is a multi-tone conversion unit, 30 is a central tone region defining unit, 40 is a tone-lowering unit, 50 is a threshold value calculating unit, and 60 is an arteriovenous device. Identification image generating means, 65 is an arteriovenous diameter ratio calculating means, 67 is an arteriovenous identification image correcting means, 70 is a fundus image storage means, and 80 is an arteriovenous identification image storage means. The multi-gradation hue conversion unit 20, the central gradation region definition unit 30, and the gradation reduction unit 40 collectively function as a hue conversion unit.

  Blood vessel region enhancement means 10, multi-gradation hue conversion means 20, central gradation area definition means 30, lower gradation reduction means 40, threshold value calculation means 50, arteriovenous identification image generation means 60, arteriovenous diameter ratio calculation The means 65 and the arteriovenous identification image correcting means 67 are realized by the CPU 1 executing a program stored in the storage device 3. The fundus image storage unit 70 is a storage unit that stores a full-color fundus image to be extracted from a blood vessel region, which is captured in full color using a visible light / light source type fundus camera, and is realized by the storage device 3. When the full-color fundus image is read by the fundus image processing apparatus and processed as it is, the RAM 2 serves as the fundus image storage unit 70. A full-color fundus image is image data recorded with three RGB components, and is obtained by photographing the fundus of a subject. In the present embodiment, RGB recorded with 8-bit 256 gradations for each color is used.

  The arteriovenous identification image storage unit 80 is a storage unit that stores the arteriovenous identification image generated by the arteriovenous identification image generation unit 60 or corrected by the arteriovenous identification image correction unit 67, and is realized by the storage device 3. Is done. The arteriovenous identification image is an image in which a pixel that is likely to be an artery has a large positive value and a pixel that is likely to be a vein has a small negative value. In this embodiment, 8-bit 256th floor This is a gray scale format (monochrome format) in a key format, and each pixel is recorded in a format having a positive / negative sign indicating whether the pixel is an artery / vein.

  Each component shown in FIG. 2 is actually realized by installing a dedicated program in hardware such as a computer and its peripheral devices as shown in FIG. That is, the computer executes the contents of each means according to a dedicated program. Note that in this specification, a computer means an apparatus having an arithmetic processing unit such as a CPU and capable of data processing, and includes not only a general-purpose computer such as a personal computer but also a portable terminal such as a tablet.

  A dedicated program for operating the CPU 1 and causing the computer to function as a fundus image processing apparatus is installed in the storage device 3 illustrated in FIG. By executing this dedicated program, the CPU 1 causes the blood vessel region enhancement means 10, the multi-gradation hue conversion means 20, the central gradation area definition means 30, the gradation reduction means 40, the threshold value calculation means 50, The function as the arteriovenous identification image generating means 60 is realized. The storage device 3 not only functions as the fundus image storage unit 70 and the arteriovenous identification image storage unit 80 but also stores various data necessary for processing as the fundus image processing apparatus.

<2. Processing action>
<2.1. Pretreatment>
First, a full-color fundus image to be processed is prepared. As a full-color fundus image, if there is an image file taken in full color by a digital fundus camera, it can be used as it is. In addition, if it is an old one recorded on a photographic medium by an analog fundus camera, the stored digital color negative / positive film, photographic paper, instant photo, etc. are read in full color with a scanner. Obtain a full-color fundus image file. Generally, a full-color fundus image is obtained by photographing in full color using a visible light / light source type fundus camera. The acquired full-color fundus image is stored in the fundus image storage means 70 of the fundus image processing apparatus. In this embodiment, R, G, B component 8-bit color images with 256 gradations are prepared as full-color fundus images.

<2.2. Process Overview>
Next, the processing operation of the fundus image processing apparatus shown in FIGS. 1 and 2 will be described. FIG. 3 is a flowchart showing an outline of processing of the fundus image processing apparatus according to the embodiment of the present invention. As described above, the full-color fundus image to be processed is 8-bit 256-gradation image data for each RGB color. Therefore, a full-color fundus image having the number of pixels Xs in the x direction and the number of pixels Ys in the y direction is represented by Image (x, y) when variables c = 0 (Red), 1 (Green), and 2 (Blue) indicating color components are used. , C) = 0 to 255 (x = 0,..., Xs-1; y = 0,..., Ys-1; c = 0, 1, 2).

First, the blood vessel region emphasizing unit 10 performs processing for enhancing the blood vessel region on the full-color fundus image to obtain a blood vessel emphasized image (step S100). Details of the blood vessel region emphasizing unit 10 and the blood vessel region emphasizing process in step S100 performed by the blood vessel region emphasizing unit 10 will be described later. The blood vessel enhancement image Ray (x, y) is a grayscale image having a blood vessel extraction degree of 256 levels indicating the probability of being a blood vessel region. In the blood vessel-enhanced image, when Tray (x, y) = 0, the non-blood vessel region is completely indicated, and when Tray (x, y) = 255, the blood vessel region is completely indicated. FIG. 4 is a diagram illustrating an example of a full-color fundus image and a blood vessel enhancement image. By executing the process of step S100 on the full-color fundus image shown in FIG. 4A, the blood vessel emphasized image shown in FIG. 4B is obtained. As shown in FIG. 4B, in the blood vessel region enhancement processing used in the present application, a pixel region determined to be a blood vessel region having a relatively large value of Tray (x, y) includes the fundus Pixels in the non-fundus region that should otherwise be included in the non-blood vessel region, such as a frame line at the time of photographing, are also included.

  If the blood vessel emphasized image is obtained, then the multi-tone hue conversion means 20 performs the hue conversion of the blood vessel region pixels (step S200). Specifically, first, for all pixels Image (x, y, c) corresponding to coordinates (x, y) satisfying the blood vessel emphasized image Ray (x, y)> 128, the following [Equation 1] The hue value h is obtained by executing the processing according to the above. In [Formula 1], r = Image (x, y, 0), g = Image (x, y, 1), and b = Image (x, y, 2).

[Formula 1]
In the case of normal eyes, when 2r> g + b, h = (g−b) α / (2r−g−b)
When 2r ≦ g + b, h = 0
For cataractous eyes When 2r and g + b are not equal, h = (g−b) α / (2r−g−b)
When 2r = g + b, h = 0

  When executing the process according to [Equation 1], the process corresponding to the normal eye and the cataract eye is performed according to the prior setting. The prior setting is performed by the operator specifying whether the eye is a normal eye or a cataract eye via the instruction input I / F 4. As [alpha] in [Formula 1], an arbitrary real value can be set, but in this embodiment, [alpha] = 768.0 is set. In the case of a normal eye, 2r> g + b is always satisfied in the blood vessel region, and pixels that do not satisfy the condition can be determined as a non-fundus region (such as a frame line at the time of photographing). However, in the case of a cataractous eye, There is a case where a certain value of b becomes large, and 2r <g + b in the blood vessel region. When the normal eye setting is used, h = 0 is erroneously determined for the blood vessel region as a non-fundus region, and many blood vessel regions are not extracted. Therefore, when it is considered to be a cataract, all the pixels are regarded as a blood vessel region except for a pixel where 2r = g + b. Although Formula 1 is applied to pixels in the blood vessel region that satisfies the blood vessel emphasized image Ray (x, y)> 128, pixels in the non-fundus region included in the fundus image, such as a frame line at the time of photographing with the fundus camera. Since it also satisfies Ray (x, y)> 128, it is processed as a blood vessel region as it is.

  The processing in [Equation 1] uses h = (g−b) α / (2r−g−b). When 2r−g−b = 0, h = 0 and 2r−g−b <0. In normal eyes, the pixel is determined to be a non-fundus region (non-blood vessel region) and h = 0, and in a cataract eye, there is a pixel satisfying 2r−g−b <0 even in the vascular region. Since it cannot be determined by the hue value whether the pixel is a region (non-vascular region) pixel, it is regarded as a pixel of the vascular region including pixels of the non-fundus region (non-vascular region), and h = (g−b) α / (2r The hue value calculated in -gb) is given.

  Subsequently, by using the calculated minimum value hmin of h and the maximum value hmax of h, a process according to the following [Equation 2] is executed to obtain a K gradation converted image Hueh (x, y). Since the K gradation conversion image Hueh (x, y) has relatively many gradations than the L gradation conversion image Hue (x, y) described later, it can also be called a multi gradation hue conversion image.

[Formula 2]
Hueh (x, y) = (h−hmin) (K−1) / (hmax−hmin)
0 ≦ Hueh (x, y) ≦ K−1

  Since Hueh (x, y) obtained as a result of the first formula of [Formula 2] is a real value, one digit after the decimal point is rounded off and replaced with an integer value. An arbitrary value can be set as the gradation K in [Formula 2], but in this embodiment, K = 2048 is set. In this case, the K gradation conversion image Hueh (x, y) is a 2048 gradation gray scale image having a value of 0-2047. Since the hue value h is obtained only for a pixel that is highly likely to be a blood vessel region where the blood vessel emphasized image Tray (x, y)> 128, the K gradation converted image Hueh (x, y) is also a blood vessel region. It becomes an image corresponding to a pixel having a high possibility of being.

  If the K gradation conversion image is obtained, then the central gradation area defining means 30 extracts the central gradation area (step S300). Specifically, first, the histogram histh (h1) (0 ≦ h1 ≦) in the pixel (x, y) satisfying the blood vessel emphasized image Ray (x, y)> 128 of the K gradation conversion image Hueh (x, y). K-1) is obtained. When the histogram histh (h1) is obtained, the median (50% position) hc of the histogram histh (h1) is obtained by executing processing according to the following [Equation 3].

[Formula 3]
Ct = Σ _{h1 = 0, K} histh (h1)
Σ _{h1 = 0, hc} histh (h1) ≧ Ct / 2

  In [Formula 3], Ct is the total number of pixels of the K gradation conversion image Hueh (x, y). The value hc of h1 that first satisfies the condition of the second equation of [Equation 3] is set as the median value.

  When the median value hc is obtained, the minimum value cmin and the maximum value cmax of the central gradation region are obtained by executing processing according to the following [Equation 4].

[Formula 4]
In the range of hc-1 ≧ h1 ≧ 0, Σ _{h1 = cmin, hc-1} histh (h1) ≧ Ct / 4
In the range of hc ≦ h1 ≦ K−1, Σ _{h1 = hc, cmax} histh (h1) ≧ Ct / 4

  In [Expression 4], the first expression is integrated in order from h1 = hc−1 to the smaller one, and the value of h1 when the first expression is established for the first time is set as the minimum value cmin, and the second expression is expressed as h1 = hc. The values are accumulated in order from the largest to the largest, and the value of h1 when the second formula is satisfied for the first time is taken as the maximum value cmax. Thereby, a range in which Ct / 4 pixels are included on both sides of the median value hc can be defined as a central gradation region.

  FIG. 5 is a diagram showing a high-accuracy histogram of 2048 gradation hues when K = 2048. FIG. 5 shows an example in which the median value hc = 1032, the minimum value cmin = 1023, and the maximum value cmax = 1037. FIG. 6 is a diagram showing a central gradation region extracted based on the histogram of FIG. In the example of FIG. 6, cmin (1023) to cmax (1037) are extracted as the central gradation region.

  Once the central gradation area is extracted, the gradation reduction means 40 performs gradation reduction of the K gradation conversion image Hueh (x, y) (step S400). Specifically, an L gradation hue conversion image Hue (x, y) is obtained by executing processing according to the following [Equation 5]. L is a positive integer.

[Formula 5]
For (x, y) that satisfies (Ray (x, y)> 128) Hue (x, y) = (Hueh (x, y) −cmin) (L−1) / (cmax−cmin)
If Hue (x, y) <0, Hue (x, y) = 0
If Hue (x, y)> L-1, then Hue (x, y) = L-1.

  As shown in [Formula 5], when the calculated Hue (x, y) exceeds a negative value or exceeds L−1, it is saturated to the minimum value 0 and the maximum value L−1, respectively. An arbitrary value can be set as the gradation L in [Equation 5], but in this embodiment, L = 256 is set. In this case, the L gradation conversion image Hue (x, y) is a 256 gradation gray scale image having a value of 0 to 255.

  FIG. 7 is a diagram illustrating an example of the L gradation conversion image Hue (x, y) when L = 256. Pixels corresponding to arteries have high luminance, and pixels corresponding to veins have low luminance. As described above, the pixels satisfying Tray (x, y)> 128 include pixels in the non-fundus region included in the fundus image, such as a frame line at the time of photographing with the fundus camera. The value of the converted image Hue (x, y) is 0 (in the case of normal eyes) or L-1 (in the case of cataract eyes) of the saturated pixel value.

  If an L gradation conversion image is obtained, then the threshold value calculation means 50 calculates a threshold value (step S500). Specifically, first, a grayscale L gradation conversion image Hue (x, y) (x = 0,..., Xs-1; y = 0,... Having a value of 0 to L−1. , Ys−1), a density histogram Hist (va) (va = 0,..., L−1) is calculated for (x, y) that satisfies Ray (x, y)> 128. Then, the boundary (threshold value) for dividing the histogram into two is set to s (s = 2,..., L−2), processing according to the following [Equation 6] is executed, and the discriminant value D (s) Is calculated.

[Formula 6]
Sum1 (s) = Σ _{va = 1, s-1} Hist (va) va
Sum2 (s) = Σ _{va = s, L-2} Hist (va) va
N1 (s) = Σ _{va = 1, s-1} Hist (va)
N2 (s) = Σ _{va = s, L-2} Hist (va)
V1 (s) = Sum1 (s) / N1 (s)
V2 (s) = Sum2 (s) / N2 (s)
W (s) = (V1-V2)
D (s) = N1 (s) N2 (s) W (s) ²

  In the first formula of the above [Formula 6], the subscript “va = 1, s−1” of Σ indicates that the sum of va from 1 to s−1 is obtained. Similarly, in the second to fourth formulas, the sum is obtained within the range of subscripts. When va = 0, it is clear that it is a vein, and when va = L−1, it is clear that it is an artery, and the pixel of the non-fundus region included in the blood vessel region is also va = 0 (normal eye Case) or va = L−1 (in the case of cataract eyes), it is excluded from the calculation range. In the above [Expression 6], N1 (s) is the number of pixels on the lower gradation side than the gradation value (pixel value) s, and N2 (s) is a pixel on the higher gradation side than the gradation value (pixel value) s. The number, V1 (s) is the gradation average value on the lower gradation side than the gradation value (pixel value) s, and V2 (s) is the gradation average value on the higher gradation side than the gradation value (pixel value) s. . For example, when L = 256, the discriminant analysis process is executed using the range of va = 1 to 254.

  D (s) is calculated according to the discriminant that is the last equation of the above [Formula 6], and s that maximizes the discriminant value D (s) is selected as the threshold value Sav for binarization (arterial vein identification). . The process of specifying the threshold value Sav in step S500 is substantially the same as a known discriminant analysis method.

  FIG. 8 is a diagram showing a 256 tone hue histogram when L = 256. The example of FIG. 8 shows an example in which the threshold value Sav = 201 is specified when the median value hc = 1032, the minimum value cmin = 1023, and the maximum value cmax = 1037.

  If the threshold value Sav is obtained, then the arteriovenous identification image generating means 60 generates an arteriovenous identification image (step S600). Specifically, first, an arteriovenous identification image Hues (x, y) is obtained by executing processing according to the following [Equation 7].

[Formula 7]
When Tray (x, y)> 128, h2 = (Hue (x, y) −Sav) × 2
H2 = 0 when Tray (x, y) ≦ 128
When | h2 |> 255, | h2 | ← 255 (the sign is the same)
Hues (x, y) = h2 (with positive and negative signs)

  In [Formula 7], Hues (x, y) is obtained based on the comparison result between the L gradation converted image Hue (x, y) and the threshold value Sav. As a result of the processing according to [Formula 7], a pixel (x, y) that is likely to be an artery has a positive value, and a pixel (x, y) that is likely to be a vein has a negative value. An arteriovenous identification image Hues (x, y) is obtained. In other words, an artery or vein attribute is given to each pixel of the arteriovenous identification image Hues (x, y) by a positive or negative sign.

  The arteriovenous identification image generation means 60 further obtains an arteriovenous determination result image AVsep (x, y, c) by executing processing according to the following [Equation 8].

[Formula 8]
AVsep (x, y, 0) = (Hues (x, y) / 2 + L / 2) × Tray (x, y) / (L−1)
AVsep (x, y, 1) = Tray (x, y) / 4
AVsep (x, y, 2) = (− Hues (x, y) / 2 + L / 2) × Tray2 (x, y) / (L−1)

  In the arteriovenous determination result image AVsep (x, y, c) as well as the full-color fundus image Image (x, y, c), variables c = 0 (Red), 1 (Green), 2 (Blue) indicating color components ). In the case of arteries, the Red component increases and the Blue component decreases, and in the case of veins, the Blue component increases and the Red component increases, and the Green component increases in the blood vessel region. However, in order to facilitate visual discrimination between the arteries and veins, the pixel values of the blood vessel emphasized image are greatly reduced. In the obtained arteriovenous determination result image AVsep (x, y, c), when L = 256, it can be displayed as a general full-color image, and the distinction between an artery and a vein can be visually grasped. .

  Once the arteriovenous identification image Hues (x, y) is obtained, the arterial / venous diameter ratio calculating unit 65 then performs the region of the original fundus image corresponding to the arterial region and the vein region identified in the arteriovenous identification image. , The arteriovenous diameter ratio is calculated by a known method (step S700).

  In the above embodiment, the arteriovenous identification image Hues (x, y) calculated by the arteriovenous identification image generating means 60 according to [Equation 7] is used to calculate the arteriovenous diameter ratio as it is. In the method of binary determination of the arteriovenous only with a single threshold value Sav with respect to the hue value of the image Hue (x, y), there are not a few pixels in which the determination of the arteriovenous is erroneous. A method of correcting a pixel to which an incorrect arteriovenous attribute is given in an arteriovenous identification image Hues (x, y) and calculating an arteriovenous diameter ratio using the corrected arteriovenous identification image Hues (x, y) However, highly accurate results can be obtained. Such correction is performed by the arteriovenous identification image correction means 67. First, the principle of the proposed correction method will be described. Blood flowing through the artery mainly flows red blood cells with oxygen-bound hemoglobin, so the hue value is inclined to red, while blood flowing through the veins flows mainly through red blood cells with hemoglobin bound with carbon dioxide, so the hue value is blue. However, since venous blood flows partially in the artery and vice versa, it is not possible to completely distinguish the arteries and veins from the hue value alone. As a method for correcting pixels in which the determination of arteriovenous is wrong, it is conceivable to utilize the consistency of blood vessel attributes. The attribute of an artery or vein in a region that can be regarded as a series of blood vessels cannot be partially reversed, so if there is a place where the attribute of the artery and vein is locally reversed, correct it as an error. Can do. However, since blood vessels branch or intersect, it is difficult to extract a region that can be regarded as a series of blood vessels from a two-dimensional fundus image. Therefore, in this application, paying attention to connected pixel blocks in a blood vessel region that can be regarded as a part of a series of blood vessels, two methods for correcting pixels in which the attributes of arteries and veins are locally inverted in the block are proposed. To do. Hereinafter, two methods of correction performed by the arteriovenous identification image correcting unit 67 will be described.

  The first method will be described. In the first method, after generating the arteriovenous identification image Hues (x, y) in step S600, the arterial and vein identification image correcting means 67 first sets k to L-2 (maximum value (L-1) -1). The pixel (x, y) satisfying the absolute value | Hues (x, y) | = k of the signed hue value is sequentially searched as a correction target pixel. Signed hues of 8 neighboring pixels (x + u, y + v) (excluding u = -1, 0, 1; v = -1, 0, 1; u = v = 0) of the correction target pixel (x, y) The processing according to the following [Equation 9] is executed on the value to calculate the positive sign hue value sum Cp and the negative sign hue value sum Cn.

[Formula 9]
_{Cp = Σ u = -1,1 Σ v} = -1,1; Hues (x + u, y + v)> 0 Hues (x + u, y + v)
Cn = −Σu _{= -1,1} Σv _{= -1,1; Hues (x + u, y + v) <0} Hues (x + u, y + v)

  In [Equation 9], the first expression is for calculating the sum of hue values that take a positive value among the eight neighboring pixels, and the second expression is a negative value among the eight neighboring pixels. This is for calculating the total sum of the hue value absolute values. Subsequently, Hue (x + um, y + vm) having the maximum absolute value among the eight neighboring pixels (x + u, y + v) is set to Hues (x + um, y + vm), and processing according to the following [Equation 10] is executed to correct the corrected arteriovenous vein An identification image Hues (x, y) is obtained.

[Formula 10]
| Hues (x + um, y + vm) |> | Hues (x, y) | is satisfied (first condition),
When Hues (x + um, y + vm)> 0 and Cp> Cn (second condition), or Hues (x + um, y + vm) <0 and Cp <Cn (third condition) are satisfied,
Hues (x, y) <-Hues (x + um, y + vm)

In [Formula 10], when the first condition is satisfied and the second condition or the third condition is satisfied, the signed hue value Hues (x, y) is replaced with the value of Hues (x + um, y + vm). Become. “Hues (x + um, y + vm)> 0” in the second condition indicates that the pixel attribute is an artery, and “Hues (x + um, y + vm) <0” in the third condition indicates that the pixel attribute is a vein. Show. The first condition is that the absolute value | Hues (x, y) | of the pixel value of the searched correction target pixel is ZT, and the maximum absolute value of the pixel value among the eight neighboring pixels of the correction target pixel | If Hues (x + um, y + vm) | is ZM, it can also be expressed as ZM> ZT. If the processing according to [Equation 10] is executed for the searched correction target pixel (x, y), the absolute value of the signed hue value searched with the same value of k | Hues (x, y) Similarly, the processing according to [Equation 9] and [Equation 10] is executed for other pixels (x, y) satisfying | = k. When the above processing is executed for all pixels (x, y) satisfying | Hues (x, y) | = k, the value of k is decremented, and pixels satisfying | Hues (x, y) | = k And the above-described processing is executed for all pixels (x, y) that satisfy | Hues (x, y) | = k. As a result of decrementing the value of k, when k = 0, the processing based on the first technique by the arteriovenous identification image correcting unit 67 is ended.
FIG. 9 is a diagram illustrating an example of correction of the arteriovenous identification image Hues (x, y) by the first method. FIG. 9A shows an example of a pixel searched under the condition of k = 70 in the arteriovenous identification image Hues (x, y) and 8 neighborhoods of the pixel. Of the nine pixels shown in FIGS. 9A and 9B, the pixel shown in the center is the pixel (x, y) to be corrected, and the other pixels are eight neighboring pixels (x + u, y + v). (Excluding u = -1, 0, 1; v = -1, 0, 1; u = v = 0). In the example of FIG. 9A, when the processing according to [Formula 9] is executed, Cp = 60 + 180 + 130 + 110 = 380, Cn = − (− 20−60−40−30) = 150, and the absolute value is maximum. The hue value Hues (x + um, y + vm) = 130.

  In the case of FIG. 9A, since the first condition of [Equation 10] is | 130 |> | −70 |, this condition is satisfied, and the second condition is 130> 0 and 380> 150. Meet the conditions. Therefore, as shown in FIG. 9B, the value of (x, y) in the center pixel is replaced with 130 in accordance with [Formula 10]. The same processing as converting from FIG. 9A to FIG. 9B is executed for other pixels searched under the condition of k = 70, and the value of k is changed from 254 to 1. This is executed for all the pixels searched by the condition.

  The second method will be described. The second method is substantially a morphological expansion process. In the second method, after the generation of the arteriovenous identification image Hues (x, y) in step S600, the arteriovenous identification image correcting unit 67 first determines the morphology of the arteriovenous identification image having a value of −255 to 255. Assuming that the size is (2N + 1) × (2N + 1), (2N + 1) × (2N + 1) pixels (2N + 1) × (2N + 1) pixels around each pixel (x, y) where N ≦ x ≦ Xs−N−1 and N ≦ y ≦ Ys−N−1. N ≧ 1) pixel blocks are sequentially extracted. As the value of N, when the size of the fundus image is 1280 × 1024 pixels, N = 1 (3 × 3 pixels) is not effective as in the first method, and therefore N = 3 (7 × 7 pixels) or more. It is desirable to set to. Then, the processing according to the following [Equation 11] similar to [Equation 9] is executed for each pixel block, and the sum Cpa of hue values having a positive sign hue value and a hue value having a negative sign are obtained. The sum total Cna of absolute values of hue values is counted.

[Formula 11]
Cpa = Σu = _{−N, N} Σv = _{−N, N; Hues (x + u, y + v)> 0} Hues (x + u, y + v)
Cna = −Σu _{= −N, N} Σv _{= −N, N; Hues (x + u, y + v) <0} Hues (x + u, y + v)

  When Cpa> Cna, the sign of the hue value Hues (x, y) of the center pixel is made positive. When Cpa <Cna, the sign of the hue value Hues (x, y) of the center pixel is negative. In either case, the absolute value is not changed. In this way, for all (Xs−2N) × (Ys−2N) pixels in the range from (N, N) to (Xs−N−1, Ys−N−1), the peripheral ( Similar correction processing is repeated while referring to 2N + 1) × (2N + 1) pixel blocks. At this time, the pixel block around the adjacent pixel includes a pixel that has already been corrected. In this application, the updated hue value is used (there is a method using the hue value before update, The correction effect is stronger than the method using the value). Further, since the correction effect is often insufficient even when the correction processing is completed for all the (Xs−2N) × (Ys−2N) pixels, (Xs−2N) × (Ys−2N). It is desirable to repeat the correction process for all the pixels a plurality of times. In this application, it is repeated 6 times.

  FIG. 10 is a diagram illustrating an example of correction of the arteriovenous identification image Hues (x, y) by the second method. Of the 25 pixels shown in FIGS. 10A and 10B, the center pixel is (x, y), and the surrounding 24 pixels are (x + u, y + v) (−N ≦ u ≦ N, −N). ≦ v ≦ N; N = 2). In the example of FIG. 10A, Cpa = 1350 and Cna = 470 according to [Formula 11].

  In the case of FIG. 10A, since Cpa (= 1350)> Cna (= 470), as shown in FIG. 10B, the hue value Hues (x, y) of 5 × 5 central pixels. Will be converted to a positive sign.

  The arteriovenous determination result image AVsep (x, y, c) generated by the arteriovenous identification image generating unit 60 or corrected by the arteriovenous identification image correcting unit 67 can be displayed on the display unit 6.

  In addition, the arteriovenous identification image generating means 60 can generate two types of gray scale arteriovenous determination result images from the full-color arteriovenous determination result image AVsep (x, y, c). Specifically, processing according to the following [Equation 12] is executed to generate a grayscale arteriovenous determination result image.

[Formula 12]
av = AVsep (x, y, 1) × 4
When AVsep (x, y, 0)> AVsep (x, y, 2),
h3 = (AVsep (x, y, 0) × 255 / av−128) × 2
When AVsep (x, y, 0) <AVsep (x, y, 2),
h3 = − (AVsep (x, y, 2) × 255 / av−128) × 2
When AVsep (x, y, 0) = AVsep (x, y, 2), h3 = 0
When h3 <−255, h3 ← −255
If h3> 255, h3 ← 255

  When displaying an arterial image in gray scale, R = G = B = h3 × av / 128, and when displaying a vein image in gray scale, R = G = B = −h3 × av / 128, which is identical to RGB. Display with gradation. However, the range of 0 ≦ R, G, B ≦ 255 should not be exceeded.

  FIGS. 11 to 13 are diagrams illustrating an example in which the arteriovenous determination result image AVsep is generated based on the above-described series of processing for the full-color fundus image of the normal eye illustrated in FIG. FIG. 11 shows the case where the correction is made by the arteriovenous identification image correction unit 67 and is generated by the arteriovenous identification image generation unit 60, and FIG. FIG. 13 shows a case where correction is performed by the arteriovenous identification image correcting means 67 using the first method and the second method. In the second method, N = 3 is set, and a series of correction processing for 694 × 599 pixels (original image: 700 × 605 pixels) is repeated six times. In each of FIGS. 11 to 13, the left side is a full color display, the upper right side is a gray scale display of an artery, and the lower right side is a gray scale display of a vein. The present invention also supports cataract eyes as described above. FIG. 14 is a diagram illustrating another example of a full-color fundus image of a cataract eye and a blood vessel emphasized image corresponding to FIG. FIGS. 15 and 16 are diagrams illustrating an example in which the arteriovenous determination result image AVsep is generated based on the above-described series of processing for the 700 × 605 pixel full-color fundus image of the cataract eye illustrated in FIG. is there. FIG. 15 shows a case where the blood vessel emphasis image shown in FIG. 14B is generated by the arteriovenous identification image generating unit 60 and not corrected by the arterial and vein identification image correcting unit 67, and is not shown. A 256-gradation conversion image similar to that in FIG. 7 is created, the threshold value Sav = 89 is calculated by the above-described discriminant analysis method, and an arteriovenous determination result image AVsep is generated. Compared to FIG. 11 of the normal eye, it is rich in blue components, and thus it can be seen that the value of Sav is remarkably low and tilted toward the vein side. Further, in FIG. 11 for the normal eye, the pixel value of the non-fundus region such as a frame line included in the blood vessel region has a value of −255 on the vein side, but in FIG. 15 for the cataract eye, +255 on the arterial side. The value will increase. FIG. 16 shows a case where correction is performed using the first method and the second method by the arteriovenous identification image correcting unit 67 as in FIG. 13. In the second method, N = 3 is similarly set, and a series of correction processing for 694 × 599 pixels (original image: 700 × 605 pixels) is repeated six times. In both of FIGS. 15 and 16, the left figure is a full color display, the upper right side is a gray scale display of an artery, and the lower right side is a gray scale display of a vein.

  Further, the arteriovenous identification image generating means 60 may interactively correct the full-color arteriovenous determination result image AVsep (x, y, c) based on an external instruction via the instruction input I / F. it can. Specifically, when the user has instructed to change the attribute of the correction region (x1 ≦ x ≦ x2, y1 ≦ y ≦ y2) and the artery or vein, all the attributes of the target pixels of the blood vessel region in the region Or change to veins. More specifically, when the target pixel is AVsep (x, y, 0)> AVsep (x, y, 2) and the attribute is changed to the artery, it is not changed as it is. When the target pixel is AVsep (x, y, 0)> AVsep (x, y, 2) and the attribute is changed to a vein, the values of AVsep (x, y, 0) and AVsep (x, y, 2) are set. Reverse each other. When the target pixel is AVsep (x, y, 0) <AVsep (x, y, 2) and the attribute is changed to an artery, the values of AVsep (x, y, 0) and AVsep (x, y, 2) are set. Reverse each other. When the target pixel is AVsep (x, y, 0) <AVsep (x, y, 2) and the attribute is changed to the vein, it is not changed as it is.

<3. Blood vessel region enhancement means>
Next, details of the blood vessel region emphasizing unit 10 and the blood vessel region emphasizing process in step S100 performed by the blood vessel region emphasizing unit 10 will be described. FIG. 17 is a functional block diagram showing details of the blood vessel region enhancement means 10. In FIG. 17, 11 is a gray scale conversion means, 12 is an image flattening means, 13 is a linear component enhancement means, and 14 is a pixel gradation conversion means.

  FIG. 18 is a flowchart showing details of the blood vessel region enhancement processing in step S100 by the blood vessel region enhancement means 10. The gray scale conversion means 11 performs gray scale conversion on the full-color fundus image Image (x, y, c) by executing processing according to the following [Equation 13] (step S110).

[Formula 13]
Gr (x, y) = Image (x, y, 1)

  In the above [Equation 13], Image (x, y, 1) represents a G (green, green) component of the full-color fundus image. Therefore, in the above [Equation 13], only the G component of each pixel is extracted to obtain a grayscale image Gr (x, y). In general, in a fundus image of a healthy person, blood vessel images appear mainly in the G component and the B component. On the other hand, in the fundus image of a patient with advanced cataract symptoms, the B component blood vessel image is negative-positive-inverted compared to a normal subject (the brightness of the B component in the background portion is improved with respect to the blood vessel region) or becomes blurred. . Therefore, by extracting the G component that is a single color component, it becomes easier to extract the blood vessel components of both healthy subjects and cataract patients.

  Further, the gray scale conversion unit 11 performs a process according to the following [Equation 14] on the gray scale image Gr (x, y), thereby performing negative / positive inversion, and gray scale fundus image Gray (x , Y). Here, the reason for negative / positive inversion with respect to the fundus image is that the luminance of the blood vessel candidate region is increased and the luminance of the non-blood vessel candidate region of the background is decreased, so that the blood vessel candidate region of the grayscale fundus image is pseudo-colored. This is because coloring can be performed, and even when a plurality of gray scale fundus images are combined, the background portion (non-blood vessel candidate region) is not mixed, and the positional relationship between the blood vessel candidate regions can be easily grasped.

[Formula 14]
Gray (x, y) = 255-Gr (x, y)

  By executing the processing according to the above [Equation 14] and performing negative / positive inversion, the pixel value of the blood vessel region having a lower luminance than the surroundings is converted into a high state. Here, negative / positive reversal means a process of converting a smaller original pixel value to a smaller value.

  When the gray scale fundus image is obtained, the image flattening unit 12 calculates the average value of the pixel values of neighboring pixels included in the predetermined range, and calculates the average value from the predetermined value to the pixel value of the pixel. A flattened image is created by adding the subtracted values (step S120). Specifically, a grayscale flattened image Gray ′ (x, y) is obtained by executing processing according to the following [Equation 15].

[Formula 15]
Mean (x, y) = [Σ _{j = −m + 1, m} Σ _{i = −m + 1, m} Gray (x + i, y + j)} / (4 m ² )
Gray ′ (x, y) = Gray (x, y) + Mid−Mean (x, y)

In the above [Expression 15], m is an integer of 1 or more, and the subscript “j = −m + 1, m” of Σ is “i = −m + 1, m” where j is from −m + 1 to m, and i is The sum of 4m ² pixels from −m + 1 to m is obtained. That is, Mean (x, y) represents an average value of pixel values of neighboring pixels included in a range of m pixels before, after, from the pixel (x, y). The number of neighboring pixels is preferably in the range of 1/3000 to 1/7000 of all pixels in the image. In the present embodiment, m = 8 and 256 (= 4 m ² = 2m × 2 m) neighboring pixels. This corresponds to about 1 / 50,000 when the gray scale fundus image Gray (x, y) is 1280 × 1024 pixels.

  Using the average value Mean (x, y), the pixel value Gray (x, y) of each pixel (x, y) is changed from the predetermined value Mid to the average value Mean (x , Y) is added to obtain a flattened image Gray ′ (x, y). The predetermined value Mid is a median value of gradations that the pixel can take. In this embodiment, since an image can take a value of 0 to 255, there are two median values of 127 and 128, but for convenience, Mid = 128.

  When calculating the average value Mean (x, y), as shown in the first equation of [Equation 15], the sum of the pixel values of neighboring pixels may be obtained directly. The processing load is high. Therefore, in the present embodiment, the image flattening unit 12 calculates the average value Mean (x, y) at high speed with a small processing load by executing processing according to the following [Equation 16]. Yes.

[Formula 16]
S (i, j) = _{Σjj = 0, j−1} Σii _{= 0, Xs−1} Gray (ii, jj) + _{Σjj = 0, i} Gray (ii, j)
Mean (x, y) = [S (x + m, y + m) −S (x−m, y + m) −S (x + m, ym) + S (x−m, ym)} / (4 m ² )

  In the first equation of [Equation 16], S (i, j) is a pixel value Gray (ii, j) from the first pixel (upper left corner of the image x = 0, y = 0) to the pixel (i, j). jj) is a cumulative value of all Xs × j + i + 1.

  FIG. 19 is a diagram illustrating a relationship between a pixel (x, y) on a grayscale fundus image and neighboring pixels for calculating an average value Mean (x, y). In order to calculate the average value Mean (x, y) as surrounded by a large thick frame, all 2m × from the upper left pixel (x−m + 1, y−m + 1) to the lower right pixel (x + m, y + m) It is necessary to add 2m pixel values. At this time, a cumulative value S (x, y) of pixel values from the first pixel of the image to the pixel (x, y) is calculated using the first equation of [Equation 16]. Then, 2m × 2m neighboring pixels and 2m pixels composed of pixels (x−m, y−m + 1) to pixels (x−m, y + m) adjacent to the left side thereof are adjacent above them. Each pixel is obtained by adding 2m pixels consisting of pixels (xm + 1, ym) to pixels (x + m, ym) and pixels (xm, ym) adjacent to the upper left of those pixels Has a cumulative value as shown in FIG.

  Then, the average value Mean (x, y) can be calculated at high speed with a small processing load by adding and subtracting the accumulated values of the four pixels, as shown in the second formula of [Formula 16]. Here, the four pixels to which the cumulative value is added are (x + m, y + m), (x−m, y + m), (x + m, ym), and (x−m) surrounded by a small thick frame in FIG. , Ym). By adding a value obtained by subtracting the average value Mean (x, y) of neighboring pixels from a predetermined value Mid to each pixel (x, y) of the grayscale fundus image Gray (x, y), a flattened image Gray ′ is obtained. (X, y) is obtained.

  Eventually, by executing the processing according to [Equation 16], the image flattening unit 12 applies (X, 0) to (Xs) in advance for the gray scale fundus image whose image size is Xs × Ys. −1, y−1) in the range of Xs × y, and x + 1 in the range of (0, y) to (x, y), the sum of all Xs × y + x + 1 pixel values is calculated for each pixel (x, A process of calculating as a cumulative value S (x, y) in y) is performed, and S (x + m, y + m) −S (x−m, y + m) −S (x + m, ym) + S (x−m, y−). A value obtained by dividing the value of m) by 2 m × 2 m is calculated as an average value. Although an extra process for calculating the accumulated value S (x, y) in each pixel (x, y) is added in advance, this calculation can be performed only by Xs × Ys addition operations. On the other hand, if the average value Mean (x, y) at each pixel (x, y) is calculated without using the cumulative value S (x, y), 2m × 2m addition is performed for each pixel. An operation is required, and a total Xs × Ys × 2m × 2m addition operation is required. On the other hand, if the cumulative value S (x, y) is used, four additions / subtractions are required for each pixel, and a total of Xs × Ys × 5 additions / subtractions including the cumulative value calculation for the entire image in advance is completed. To do.

  Once the flattened image is obtained, the linear component enhancing means 13 performs an opening process (shrinkage / expansion process) on the flattened image to emphasize the linear component, and the linear component emphasized image. Is obtained (step S130). FIG. 20 is a flowchart showing the linear component enhancement processing operation in step S130.

  First, the linear component enhancement unit 13 performs an opening process using eight types of linear structural elements on the flattened image obtained by the image flattening unit 12 (step S131). In the opening process in step S131, a plurality of linear opening images are created using a plurality of linear structural elements. In the present embodiment, eight linear structural elements are used as designated structural elements. This is a linear structural element that is a linear structural element whose direction is different in eight directions. FIG. 21 is a diagram illustrating an example of a linear structural element used in the present embodiment.

  Since the linear structural element shown in FIG. 21 serves as a mask, it is a binary image. Therefore, in FIG. 21, the value of the pixel indicated as “0” is “0”, but the value of the pixel indicated as a numerical value other than “0” is “1” according to each of the eight types of directions d. ”And“ 0 ”in the other direction. Numbers other than “0” in FIG. 21 indicate directions. “1” is the horizontal direction (the horizontal direction in the drawing), and thereafter the direction changes at intervals of 22.5 degrees as the number increases by one. “2”, “4”, “6”, and “8” are arranged in adjacent sections because columns of the same number cannot be arranged on a straight line because of the lattice arrangement of pixels. Three-digit numbers such as “234” and “678” are in the three directions “2” “3” “4” and “6” “7” “8”, respectively. It shows that the pixel value is “1”. Note that “9” indicates the center pixel, and the value of the center pixel is always “1”.

  As shown in FIG. 21, the linear structuring element is twice as long as the radius R of the circular structuring element, has a pixel width of one pixel, and is arranged at least in the vicinity of straight lines spaced at 22.5 degrees, and has eight directions. Is defined. Actually, it is preferable that the continuous pixels to which the pixel value “1” is given in each direction be a straight line, but it is not always a straight line when the total number of pixels of the linear structural element is small. Therefore, in the example of FIG. 21, in the directions indicated by “2”, “4”, “6”, and “8”, the pixel having the pixel value “1” is not in the vicinity of the straight line having an interval of 22.5 degrees. Will be placed.

  The actually used linear structural element is obtained by adding a condition of a radius of 7 pixels from the center. That is, the logical product (AND) of the linear structural element shown in FIG. 21 and the circular structural element shown in FIG. 22 is obtained. FIG. 22 is a diagram illustrating an example of a circular structural element used in the present embodiment. As shown in FIG. 22, in this embodiment, a circular structural element that is a form of a 15 × 15 mask image in which a radius of 7 pixels, that is, a distance of 7 pixels or less from the center is effective is used. As shown in FIG. 22, among the 15 × 15 pixels, the pixel value at a distance of 7 pixels or less from the center is “1”, and the pixel value of a pixel exceeding the distance 7 from the other center is “0”. By taking the logical product (AND) of the linear structuring element shown in FIG. 21 and the circular structuring element shown in FIG. 22, for example, in the case of direction d = 1, a binary mask as shown in FIG. An image is obtained. When the direction d = 2, a binary mask image as shown in FIG. 24 is obtained. In FIG. 23 and FIG. 24, a pixel indicated as “1” is a reference pixel. That is, in the binary image of (2R + 1) × (2R + 1) pixels in which the reference pixel to be referred to in the opening process is defined, the reference pixel is defined inside the circle with the radius R, and the pixel width is 2R + 1 in length. Eight directions are defined by arranging one pixel at least in the vicinity of straight lines having an interval of 22.5 degrees.

  The linear structural elements shown in FIG. 21 are M (d, u, v) = {0, 1} (d = 1,..., 8; u = −R,..., 0,. V = −R,..., 0,. Note that R indicates an effective radius, and R = 7 in the example of FIG. U represents the horizontal axis in FIG. 21, and v represents the vertical axis in FIG. The linear component emphasizing means 13 uses the linear structural element M (d, u, v) on the flattened image Gray ′ (x, y) obtained by executing the processing according to the above [Equation 15]. On the other hand, processing according to the following [Equation 17] is executed to obtain a contracted image Eray (x, y).

[Formula 17]
Eray (d, x, y) = MIN _{u = −R, R; v = −R, R} [(255−M (d, u, v) × 254) × (Gray ′ (x + u, y + v) +1) − 1]

  In the above [Equation 17], MIN indicates that it takes the minimum value, and the subscript “u = −R, R; v = −R, R” of MIN indicates that u and v are from −R to R. It shows performing an operation on a range. That is, for each pixel Gray ′ (x, y) of “x = R,..., Xs−R−1; y = R,. , R; v = −R, R ″ is converted into the minimum pixel value of M (u, v) = 1 in the adjacent pixels. Therefore, in the example of FIG. 21, the calculation is performed using 225 pixels with R = 7 for (2R + 1) × (2R + 1) pixels of the linear structure element, and the minimum value is given as Ray (x, y). .

  When the contracted image Eray (d, x, y) is obtained by executing the processing according to the above [Equation 17], this Eray (d, x, y) is replaced with Gray ′ (x, y), It is also possible to repeatedly execute the process according to [Equation 17]. The number of repetitions can be set as appropriate. However, it is normally set once and not repeated in order to suppress image quality deterioration. By performing for each of the eight directions d, eight kinds of contracted images Eray (d, x, y) of d = 1 to 8 are obtained.

  If the contraction process is repeatedly executed and eight kinds of contraction images Eray (d, x, y) are obtained, then the linear component emphasizing means 13 uses the linear structural element used for the contraction process as the designated structural element. The expansion process is performed. Specifically, using the linear structural element M (d, u, v), the contracted image Eray (d, x, y) is subjected to processing according to the following [Equation 18], and after expansion Image Dray (d, x, y) is obtained.

[Formula 18]
Dray (d, x, y) = MAX _{u = −R, R; v = −R, R} [M (d, u, v) × Eray (d, x + u, y + v)]

  In the above [Equation 18], MAX indicates the maximum value, and the suffix “u = −R, R; v = −R, R” of MAX indicates that u and v are from −R to R. It shows performing an operation on a range. That is, for each pixel Eray (d, x, y) of “x = R,..., Xs−R−1; y = R,. R, R; v = −R, R ″ is converted into the maximum pixel value of M (d, u, v) = 1 in the adjacent pixels. Therefore, in the example of FIG. 21, the calculation is performed using 225 pixels as (2R + 1) × (2R + 1) pixels of the linear structure element, and the maximum value is given as Dray (d, x, y).

  When the expanded image Dray (d, x, y) is obtained by executing the processing according to [Formula 18], this Dray (d, x, y) is replaced with Eray (d, x, y). Thus, the processing according to the above [Formula 18] can be repeatedly executed. The number of repetitions needs to be the same as the number of repetitions of the contraction process. Normally, the contraction process is set to one and no repetition is performed. By performing for each of the eight directions d, eight kinds of post-expansion images Dray (d, x, y) of d = 1 to 8 are obtained. An image obtained as a result of contraction and expansion is obtained as a linear opening image Dray (d, x, y).

  Next, the linear component emphasizing unit 13 performs a process of creating an image of the maximum value of the eight types of linear opening images (step S132). Specifically, processing according to the following [Equation 19] is executed to obtain a linear component emphasized image Lray (x, y) that is an image of the maximum value.

[Formula 19]
Lray (x, y) = MAX _{d = 1,8} Dray (d, x, y)

  In the above [Equation 19], MAX indicates that it takes the maximum value, and the subscript “d = 1, 8” of MAX is in all eight types of linear opening images Dray (d, x, y). It shows performing an operation. That is, for each pixel (x, y), a process of acquiring the maximum value of eight types of linear opening images Dray (d, x, y) is performed. As a result, a linear component emphasized image Lray (x, y) that is an image of the maximum value is obtained.

  After the linear component enhanced image is obtained, the pixel gradation converting unit 14 performs gradation conversion (contrast correction) on the linear component enhanced image to obtain a blood vessel enhanced image (step S140). Specifically, first, the pixel gradation conversion unit 14 calculates the frequency distribution of the pixel value vb for all Xs × Ys pixels of the linear component emphasized image Lray (x, y) having a value of 0 to 255. H (vb) (vb = 0,..., 255) is obtained. Then, the minimum value vbmin that satisfies the condition shown in the following [Equation 20] is obtained.

[Formula 20]
_{Σvb = 0, vbmin} H (vb) ≧ (Xs × Ys) × α

  In the above [Equation 20], the subscript “vb = 0, vbmin” of Σ indicates that the sum of vb from 0 to vbmin is obtained. Therefore, the condition shown in the above [Expression 20] indicates that, of the total number of pixels (Xs × Ys), pixels with a ratio α × 100% have pixel values equal to or less than the pixel value vbmin. As the ratio α, it is necessary to give a ratio of a region other than the blood vessel included in the fundus image, and this ratio varies for each acquired fundus image, but on average 0.7 to 0.9, typically As a specific example, it is preferable to set 0.8. Eventually, by using [Equation 20], the minimum pixel value vb of the linear component enhanced image is used as a threshold value of the pixel value for discriminating between the non-blood vessel candidate region and the blood vessel candidate region included in the linear component enhanced image. It is possible to specify the minimum pixel value vbmin among pixel values in which the total number of pixels counted from 0 exceeds a predetermined ratio α of the total number of pixels of the linear component emphasized image.

  Once the minimum value vbmin that satisfies the condition shown in [Equation 20] is obtained, the pixel gradation conversion means 14 then x = 0,..., Xs−1; y = 0,. The processing according to the following [Equation 21] is performed on all the pixels of −1 to perform contrast correction, and a blood vessel emphasized image Ray (x, y) is obtained.

[Formula 21]
Ray (x, y) = {Lray (x, y) −vbmin} × 255 × β / (255−vbmin)

  In the above [Equation 21], β is a luminance scaling value. The luminance scaling value β can be set arbitrarily, but depends on the set ratio α, so when α = 0.8, it is preferably 70 to 90, and particularly preferably 80. Can do. That is, β is preferably set to about 100 times α.

  Since Ray (x, y) is a pixel value, when the result of [Formula 21] becomes a negative value, it is replaced with Ray (x, y) = 0. Therefore, pixels that do not satisfy the predetermined condition Lray (x, y)> vbmin are replaced with the minimum value “0”, and pixels that satisfy the predetermined condition Lray (x, y)> vbmin are from the minimum value “0”. The pixel value of Tray (x, y) is set so as to be in the range from the large value “1” to the maximum value. In Tray (x, y), “0”, which is the minimum value of the gradation, indicates the outside of the blood vessel candidate region, and the blood vessel candidate region is “1” or more and “255” or less. As the minimum value of the blood vessel candidate region, a value near the minimum value of the gradation is set. In the present embodiment, the minimum gradation value + 1 is set as the value near the minimum gradation value, but it may be more than that. Of course, in order to increase the contrast of the blood vessel region, it is preferable that the minimum value of the gradation is +1. The blood vessel emphasized image Ray (x, y) obtained in this way becomes a gray scale image having a higher value as the pixel having a higher probability of being a blood vessel region.

  When a pixel that does not satisfy the predetermined condition Lray (x, y)> vbmin is set to the minimum value, it is preferably set to “0” as described above, but may be a value other than “0”. Since the fundus region is generally circular, the non-fundus region is also included in the rectangular fundus image. Therefore, the non-fundus region is set to “0” and the non-blood vessel candidate region in the fundus region is set to “1”. It is also possible to set the blood vessel candidate region in the fundus region to “2” to “255”. Since the purpose is to increase the contrast, as a matter of course, the minimum value in the blood vessel candidate region is preferably set to a value close to “0”.

  In this embodiment, since processing is performed on eight types of linear opening images according to eight types of directions, extraction of granular components can be suppressed evenly in each direction changed at an equal interval of 22.5 degrees. Since the number of images to be created is only eight according to the direction, it is possible to reduce the load of the arithmetic processing.

<4. Modified example>
The preferred embodiments of the present invention have been described above. However, the present invention is not limited to the above embodiments, and various modifications can be made. For example, in the above embodiment, a full-color image is used as a fundus image obtained by photographing. However, in recent years, the gradation of a commercial digital camera has been expanded to 10 bits or more, and a full-color image such as 1024 gradations for each color is used. It is also possible to obtain a color image having a larger number of gradations.

1 ... CPU (Central Processing Unit)
2 ... RAM (Random Access Memory)
3 ... Storage device 4 ... Instruction input I / F
5. Data input / output I / F
6 ... Display unit 10 ... Blood vessel region enhancement means 11 ... Gray scale conversion means 12 ... Image flattening means 13 ... Linear component enhancement means 14 ... Pixel gradation conversion means 20 ..Multi-tone hue conversion means (hue conversion means)
30: Median gradation area defining means (hue converting means)
40... Gradation reduction means (hue conversion means)
50 ... Threshold value calculating means 60 ... Arteriovenous identification image generating means 65 ... Arteriovenous diameter ratio calculating means 67 ... Arteriovenous identification image correcting means 70 ... Fundus image storage means 80 ... Arteriovenous identification image storage means 100 ... Fundus image processing apparatus

Claims (12)

A blood vessel region enhancement means for obtaining a grayscale blood vessel enhancement image in which a higher value is set for a pixel having a higher probability of being a blood vessel region with respect to a color fundus image;
A pixel having a value greater than or equal to a predetermined value in the blood vessel enhanced image is extracted as a pixel of the blood vessel region, and an RGB value is referred to each pixel of the blood vessel region with reference to a corresponding pixel of the color fundus image. Hue conversion means for acquiring and converting the acquired RGB value to a hue value of L gradation (L is a positive integer) and obtaining a hue conversion image in which the pixel value is converted to the hue value of L gradation;
Threshold calculation that calculates a histogram of L gradation based on pixels corresponding to the blood vessel region of the hue conversion image, and calculates a threshold value for binarization by a discriminant analysis method using the calculated histogram Means,
An arteriovenous identification image to which an attribute including at least an artery or a vein is given is obtained based on a comparison result between an L tone hue value of each pixel corresponding to the blood vessel region of the hue conversion image and the threshold value. An arteriovenous identification image generating means;
A fundus image processing apparatus comprising: The hue conversion means includes
A pixel having a value greater than or equal to a predetermined value in the blood vessel enhanced image is extracted as a pixel of the blood vessel region, and an RGB value is referred to each pixel of the blood vessel region with reference to a corresponding pixel of the color fundus image. A multi-level image obtained by converting the acquired RGB value into a hue value of K (K is an integer greater than L) gradation and obtaining a multi-gradation hue conversion image in which the pixel value is converted into a hue value of K gradation. Toning hue conversion means;
A central gradation area defining means for defining a central gradation area within a range of cmin (cmin is a positive integer) to cmax (cmax is larger than cmin and smaller than K-1) within the range of the K gradation;
With respect to the multi-tone hue conversion image, the hue conversion image of L gradation having a value from 0 to L-1 from 0 to L-1 is obtained by saturating cmin or less to 0 and cmax or more to L-1. A gradation reduction means for conversion;
The fundus image processing apparatus according to claim 1, further comprising:   The multi-gradation hue conversion unit converts the acquired RGB values into hue values for all pixels in the blood vessel region, and then sets the minimum hue value to 0 and the maximum hue value to K-1. The fundus image processing apparatus according to claim 2, wherein the fundus image processing apparatus converts the hue value into an integer hue value of K gradations.   The center gradation area defining means calculates the total number of pixels corresponding to the blood vessel area in a high-accuracy histogram calculated based on the pixels corresponding to the blood vessel area of the multi-tone hue conversion image, Ct, 4. The fundus image processing apparatus according to claim 2, wherein a median value is defined as the median gradation region, where hc is a median value and Ct / 4 pixels are included on both sides of the median value hc. . The blood vessel region enhancement means includes
Grayscale conversion means for obtaining a grayscale fundus image based on at least one of the color components of the color fundus image;
For each pixel of the grayscale fundus image, an average value of pixel values of neighboring pixels included in a predetermined range is calculated, and a value obtained by subtracting the average value from the predetermined value is added to the pixel value of the pixel. Image flattening means for creating a flattened image,
Linear component enhancement means for performing an opening process on the flattened image using a predetermined structural element and creating a linear component enhanced image in which the luminance of the blood vessel candidate region is equal to or higher than a certain level;
In the linear component emphasized image, a pixel value that does not satisfy the predetermined condition is replaced with the minimum gradation value of the converted blood vessel emphasized image, and the pixel value that satisfies the predetermined condition is the minimum gradation value. Pixel gradation conversion means for correcting the gradation of the pixel so as to be in the range of the maximum value from the value near the value, and creating a blood vessel emphasized image;
The fundus image processing apparatus according to any one of claims 1 to 4, further comprising:   The hue conversion means extracts a pixel having a value of 128 or more as a pixel of a blood vessel region when the blood vessel emphasized image has a value of 256 gradations. The fundus image processing apparatus according to one item.   The multi-gradation hue conversion means converts the acquired RGB values into K gradation hue values by setting values corresponding to R, G, and B to r, g, and b, and α is a predetermined real value. And h = (g−b) α / (2r−g−b), h = 0 when 2r−g−b = 0, and h = 2 for the normal eye when 2r−g−b <0. The fundus according to claim 2, wherein in the case of a cataract eye, conversion is performed based on the obtained value of h as h = (g−b) α / (2r−g−b). Image processing device.   The threshold value calculation means calculates the threshold value by setting the L gradation to 256 gradations, and executing the process of discriminant analysis using the range of gradation 1 to gradation 254 of the calculated histogram. The fundus image processing apparatus according to any one of claims 1 to 7, wherein the fundus image processing apparatus comprises:   The arteriovenous identification image generating means sets the L gradation to 256 gradations and the threshold value to Sav, with respect to the hue value Hue of each pixel corresponding to the blood vessel region of the hue conversion image ( Hue-Sav) × 2 is performed to convert the pixel value to a range of −255 to 255, and a pixel value of 0 is set for each pixel not corresponding to the blood vessel region. If the pixel value is negative, the pixel is assigned a vein attribute, and if the pixel value is 0, the arteriovenous identification image is such that the pixel is not included in the blood vessel region. The fundus image processing apparatus according to any one of claims 1 to 8, wherein: With respect to the arteriovenous identification image generated by the arteriovenous identification image generating means, sequentially search for pixels having pixel values from (maximum pixel value-1) to 1 as correction target pixels,
The absolute value of the pixel value of each of the correction target pixels searched for is ZT, the maximum absolute value of the pixel values among the eight neighboring pixels of the correction target pixel is ZM, and the arterial attribute is selected among the eight neighboring pixels. ZM> ZT, where Cp is the sum of absolute values of the pixel values of pixels having Cp and Cn is the sum of absolute values of the pixel values of pixels having the vein attribute among the eight neighboring pixels,
When the attribute of the pixel having the maximum value ZM satisfies Cp> Cn in the artery, or the attribute of the pixel having the maximum value ZM satisfies Cp <Cn in the vein, the attribute of the correction target pixel is set to the pixel having the maximum value ZM. An arteriovenous identification image correcting unit that sequentially corrects all of the searched correction target pixels so as to replace the attribute and replace the absolute value of the pixel value of the correction target pixel with ZM. The fundus image processing apparatus according to any one of claims 1 to 9, further comprising: A pixel block of (2N + 1) × (2N + 1) pixels (N ≧ 1) is sequentially extracted from the arteriovenous identification image generated by the arteriovenous identification image generating means, and has an arterial attribute in the pixel block. If the total number of pixels is Cpa, and the total number of pixels having the vein attribute in the pixel block is Cna,
When Cpa> Cna, the attribute of all pixels in the pixel block is set as an artery, and when Cpa <Cna, the correction is performed so that the attributes of all pixels in the pixel block are set as veins. The fundus image processing apparatus according to claim 1, further comprising a vein identification image correction unit.   A program for causing a computer to function as the fundus image processing apparatus according to any one of claims 1 to 11.