WO2007110982A1

WO2007110982A1 - Image analyzer and program for stereo eye fundus image

Info

Publication number: WO2007110982A1
Application number: PCT/JP2006/318912
Authority: WO
Inventors: Hiroshi Fujita; Toshiaki Nakagawa; Yoshinori Hayashi
Original assignee: Tak Co., Ltd.
Priority date: 2006-03-24
Filing date: 2006-09-25
Publication date: 2007-10-04
Also published as: JP2007252707A; JP5278984B2

Abstract

[PROBLEMS] To provide an image analyzer whereby the three-dimensional shape of an optic papilla can be accurately and efficiently analyzed from a stereo eye fundus image. [MEANS FOR SOLVING PROBLEMS] An image analyzer whereby the pixel data of the area other than optic papillae in a stereo image is analyzed, the whole image including the optic papilla area is positioned, the pixel data of the optic papilla area in the stereo image thus positioned is analyzed, the point photographing the same position in an optic papilla is determined, and the depth data of the optic papilla is calculated based on the position data of the corresponding point in each image.

Description

Specification

Image analysis apparatus and program for stereo images of the fundus

The present invention relates to an image analysis apparatus and an image analysis program for reproducing a three-dimensional shape of a measurement object based on a stereo image obtained by photographing different position force measurement objects. In particular, the present invention relates to an image analysis apparatus and an image analysis program for analyzing information of a stereo image obtained by photographing a fundus oculi and obtaining 3D information for accurately reproducing the 3D shape of the optic nerve head.

Background art

[0002] Fundus images obtained by photographing the fundus are widely used for eye examination and disease diagnosis. In particular, observing the shape of the optic disc from the fundus image is very useful in confirming the presence and progression of glaucoma. Therefore, there is a need for a technique that can observe the exact three-dimensional shape of the optic disc. As a means of observing the three-dimensional shape of the optic nerve head, various attempts have been made to reproduce the three-dimensional shape of the optic nerve head based on stereo images obtained by photographing the fundus from different positions. The

[0003] For the purpose of obtaining information indicating the three-dimensional shape of the fundus, FIG. 1 schematically shows a basic photographing position when a stereo image of the fundus of a subject is photographed. Photographing is performed so that the optical axes are parallel to each other at positions 22 and 24, which are the base line length. The first fundus image 26 is captured by photographing at the position 22, and the second fundus image 28 is photographed by the camera at the position 24. The focal lengths of the camera at position 22 and the camera at position 24 are adjusted so that the plane determined by the fundus image 26 and the plane defined by the fundus image 28 are the same.

[0004] A method of analyzing the image 26 and the image 28 to obtain a three-dimensional coordinate value indicating the shape of the fundus will be briefly described. In the stereo image analysis, first the point P in the first image 26 and the

Shi

The point P to be measured at the same position in real space is captured as shown in point 28 of image 28 in 2.

R

It is necessary to determine the position of the point on the image. Such a point is called a corresponding point. Usually, a pine that compares the characteristics of pixel values for each pixel to find corresponding points on the image. Ching is performed. As a result of matching, the point corresponding to the point force with the most similar characteristics of the pixel value is determined.

[0005] When the corresponding point P and the position of the point P in the image are determined, the coordinate system of the image of the point P

L R L

Coordinate value P (X, y))

The point P

Coordinate value P (X, y shown in the coordinate system of R image

R R R can be used to find the coordinates (X, y, z) of point P shown in the global coordinate system of real space. The coordinate value of the point P is expressed by the following equation (1) (2) (3), with the coordinate value P (X, y) and coordinate value P

しししし

(X, y)

It is obtained by substituting R R R and solving the equation. In particular, the value of z, which is the depth information of the measurement target point, corresponds to a corresponding point due to the camera base length L, the focal length f of the camera, and the difference in shooting position between the first image and the second image. Deviation X-X

It can be easily calculated from LR. The difference in shooting position between the first image and the second image, such as the value of X-X

L R

The amount of deviation of the corresponding points due to is referred to as parallax.

[0006] [Equation 1]

x

--

-R

L

y = y _L x • · (2)

L

z = t X

[0007] Since the stereo image is obtained through the lens of the camera, the measurement object is recorded in the image in a shape including distortion derived from the optical characteristics of the lens. For this reason, it is necessary to correct image distortion in order to accurately measure the position of corresponding points and to obtain parallax. In particular, since the fundus is photographed through a crystalline lens or cornea of the eyeball, a stereo image of the fundus is photographed with an image including distortion caused by these factors. Because there are individual differences in the shape of the lens and cornea of the eyeball, it is difficult to accurately determine the corresponding point because it is impossible to know the exact fundus shape only by correction using general optical calculations. . [0008] For this reason, various attempts have been made to determine the corresponding points by correcting the distortion included in the fundus stereo image and to determine an accurate three-dimensional shape. For example, Patent Document 1 discloses a technique for correcting the position of an image using information on red, blue, and green in the upper and lower thirds of a stereo image to determine corresponding points. Patent Document 2 discloses a technique for correcting an image by applying a quadric surface assuming that the eyeball is an elliptical spherical surface.

Patent Document 1: Japanese Patent Laid-Open No. 2000-245700

Patent Document 2: Japanese Patent Laid-Open No. 2002-34924

[0009] However, these conventional techniques perform correction by analyzing pixel data for the entire image or two-thirds of the image in order to correct the position between the images. The image processing for correction took a very long time.

[0010] As a specific example of stereo images of the fundus, the amount of light passing through the lens and cornea and the wavelength of the light vary depending on the illumination conditions and the shooting direction, and the same part of the measurement object should be shot. Pixel value force May vary depending on the image. The amount of change in pixel value is particularly large in the retinal area that is the periphery of the imaging area, and the corresponding points may not be accurately extracted in the retinal area. Then, because the 3D shape of the fundus was analyzed based on the information on such incorrect corresponding points, even if it is a smooth flat surface that does not actually have unevenness, it is recognized as a structure with large unevenness and the result is displayed. In many cases, improvements were required.

Disclosure of the invention

Problems to be solved by the invention

[0011] The present invention has been made in view of the above problems, and obtains an accurate parallax value from a stereo image of the fundus oculi and efficiently and accurately analyzes the three-dimensional shape of the optic disc. The object was to provide an image analysis device and an image analysis program.

[0012] Further, the present invention provides an image analysis device that accurately analyzes the three-dimensional shape of the optic disc even when there is a change in pixel value due to illumination conditions between stereo images of the fundus. The object is to provide an image analysis program.

Means for solving the problem [0013] The invention of claim 1 relates to an image analysis apparatus that analyzes a three-dimensional shape of the optic nerve head from a stereo image obtained by photographing the fundus of a subject. The image analysis apparatus of the present invention includes an image input means for inputting pixel data of a stereo image, an area identifying means for identifying an area of the optic nerve head and an area other than the optic nerve head for each stereo image, and an optic nerve head of a stereo image. Analyzing pixel data in other areas, what? Image correction means for aligning the entire image including the area of the optic nerve head and pixel data of the optic nerve head area of the stereo image aligned by the image correction means are analyzed, and The corresponding point determination means for determining the corresponding point by taking the same part of the image and calculating the depth information of the optic nerve head from the position information in each image of the corresponding point determined by the corresponding point determination means. Including depth information determination means.

[0014] A stereo image of the fundus mainly captures the optic disc, the retina, and a plurality of blood vessels running on the retina. As a result of various studies, the inventor has photographed a stereo image of the fundus of the region other than the optic disc, that is, the retina and a plurality of blood vessels running on the retina! By analyzing the pixel data of this area and correcting the position of the image, It was found that the three-dimensional shape of the optic nerve head can be analyzed accurately and efficiently. The optic nerve head has a conical shape with a concave center, and the apex of the cone is located on the far side of the image. The retina is curved along the shape of the eyeball, but because of its relatively large diameter, it is photographed as a substantially plane that is substantially parallel to the image. The inventor pays attention to the difference in the shape of the optic disc and the retina. By analyzing the area other than the optic nerve head, we found that the entire image can be aligned efficiently, and using the stereo image after alignment, the corresponding points of the optic nerve area are determined and analyzed. By doing so, it became possible to calculate the depth information of the optic disc accurately.

[0015] The image correction means of the invention of claim 2 evaluates a total value of differences between pixel values at the same position between stereo images, and calculates a vertical movement amount and a horizontal movement amount between images. The image position correction is performed by obtaining the enlargement / reduction ratio in the horizontal direction.

[0016] As a result of analyzing the correction method by analyzing many images, the inventor found that the correction of the position of the stereo image of the fundus occupies the vertical and horizontal movements of the image and the horizontal scaling. I found out that it would be possible. The amount of image movement and the enlargement / reduction ratio can be set between stereo images. Since the sum of the differences between the pixel values at the same position can be obtained and calculated based on this value, the correction is performed efficiently and quickly.

[0017] The corresponding point determining means according to claim 3 sets a reference point in a region of the optic disc of the reference image, sets a region of interest in a specific range around the reference point, A region having an array of pixel data that is most similar to the pixel data of the region of interest is searched from the region of the optic nerve, and pixels that are sufficiently similar to the region of interest in the region of the optic nerve of other images

In this case, the region of the optic nerve head in another image is expanded or reduced in the horizontal direction to search for a region having an array of pixel data that is most similar to the pixel data of the region of interest, and based on the search result. It is characterized by determining corresponding points between the image used as a reference and another image.

[0018] The corresponding point determining means of the present invention determines a region of interest in the optic disc region of the reference image when extracting corresponding points in the optic disc region, and from the optic disc region of other images. Then, an array of pixel data similar to the pixel data of the region of interest is searched. If an array of similar pixel data cannot be obtained, a region having a similar array of pixel data may be searched by performing lateral enlargement or reduction of the optic nerve head region of another image. it can. The image of the optic nerve head region may have a lateral distortion different from the image outside the peripheral region, but the corresponding point determination means of the present invention can also cope with this distortion. Therefore, the corresponding points can be set in the image with higher accuracy.

[0019] The invention of claim 4 relates to an image analysis program for analyzing the three-dimensional shape of the optic nerve head from pixel data of a stereo image obtained by photographing the fundus of a subject. The image analysis program of the present invention discriminates each pixel data of the region of the optic disc and the region other than the optic disc for each stereo image, analyzes the pixel data of the region other than the optic disc of the stereo image, and analyzes the optic disc of the optic disc. Align the entire image including the area, analyze the pixel data of the optic nerve head area of the aligned stereo image, Corresponding points at which the same part of the optic nerve head is photographed are determined, and the computer is caused to execute a process of calculating depth information of the optic nerve head from position information in each image of the corresponding point.

The invention's effect

[0020] Stereo obtained by photographing the fundus using the image analysis apparatus and the image analysis program of the present invention The pixel data of the image can be analyzed efficiently, and 3D information that can accurately reproduce the 3D shape of the papillary optic nerve can be obtained.

[0021] The image analysis apparatus and the image analysis program of the present invention accurately analyze the three-dimensional shape of the optic nerve head even when there is a change in pixel values due to illumination conditions between stereo images. can do.

BEST MODE FOR CARRYING OUT THE INVENTION

Hereinafter, as an example of a preferred embodiment of the image analysis apparatus of the present invention, an image analysis apparatus that reproduces a three-dimensional shape of the optic nerve head from a stereo image obtained by photographing the fundus is referred to the drawings. This will be described in detail.

[0023] The image analysis apparatus according to the present embodiment is composed of one computer. Image input means for inputting pixel data of a stereo image of the image analysis device, area identification means for identifying an area of the optic nerve head and an area other than the optic nerve head for each stereo image, and pixels of an area other than the optic nerve head of the stereo image Analyzing the data? Image correction means for aligning the entire image including the area of the optic nerve head and the pixel data of the optic nerve area of the aligned stereo image are analyzed to capture the same part in the optic nerve head. Corresponding point determining means for determining the corresponding corresponding points and depth information determining means for calculating the depth information of the optic disc from the position information in each image of the determined corresponding points are the storage unit (ROM, RAM). These means can perform analysis by appropriately calculating the pixel data of the stereo image using the CPU of the computer.

[0024] General characteristics of the shape of the optic disc and retina analyzed by the image analysis apparatus of this embodiment will be described. The retina is the innermost film of the eyeball and has a curved shape along the surface of the eyeball. The optic nerve head is located in the center of the eyeball at a distance from the fovea. It also has a conical recess in the center and the nipple marginal force around the recess. In other words, the apex of the conical recess is located on the farthest side. The size of the depression of the optic nerve head varies greatly depending on the individual and the presence or absence of disease, but generally the diameter is a few percent of the diameter of the eyeball.

[0025] Stereo images of the fundus are slightly different from image to image by passing the frontal force light of the eyeball through the eyeball. The picture is taken from the shooting position. The stereo images analyzed in this embodiment are the image 2 (hereinafter also referred to as the left image) taken from the left side of the fundus shown in FIG. 2 (a), and the right side of the fundus shown in FIG. 2 (b). It consists of two images 4 (hereinafter also referred to as the right image). In the stereo image of the fundus, the optic disc region 6, the retinal region 8, and a plurality of blood vessels 10 running on the retina are photographed.

[0026] Stereo images 2 and 4 of the fundus are taken in this embodiment so that the optic disc region 6 is arranged at approximately the center of the image. In the stereo image of FIG. 2, the nipple edge 12 in the optic nerve head region 6 is photographed in a bright orange color, and the depressed portion 14 is brighter and lighter than the nipple edge 12 in orange. Taken in color. The retinal region 8 that is captured in the brown image to the dark brown color tone in the stereo image in Fig. 2 is actually curved to be convex in the depth direction with respect to the image, but the curvature of the curve is large. Since the shooting range is relatively narrow, the amount of change in the depth direction in the shooting range can be handled as a plane that is almost parallel to the image plane with very little.

[0027] Fig. 3 shows the flow of analysis performed by the image analyzer to analyze the three-dimensional shape of the optic nerve head region 6 from the two stereo images 2, 4 of the fundus. Will be explained. The image input means of the image analyzer first inputs pixel data of two stereo images 2 and 4 in step s2. The pixel data of the stereo image in this embodiment includes R (red), G (green), and B (blue) data for each pixel.

[0028] Next, in step s4, the image analysis device extracts an outline of the optic disc region 6 for each stereo image using the region identification means. In the stereo images 2 and 4, the luminance value calculated based on the R, G, and B values of the nipple edge region 12 of the optic nerve region 6 is higher than that of the retinal region 8. or,? In the part where the optic disc region 6 and the retinal region 8 are adjacent to each other, the boundary (hereinafter also referred to as an edge) where the change in luminance value is large is clear. Therefore, by evaluating the pixel data for each pixel in the region with a large change in luminance value, selecting a pixel having a predetermined threshold value or higher and connecting the closed columns of the pixels, the optic nerve head is connected. The contour of area 6 can be determined. The contour can be determined with high accuracy by applying the dynamic contour extraction method.

[0029] The region identifying means extracts the contour of the optic disc region 6 by the dynamic contour extraction method. The method will be described in more detail. First, as preprocessing, a process of temporarily deleting a plurality of blood vessels 10 captured in the stereo images 2 and 4 from the image is performed. This is a process for avoiding erroneous extraction of the boundary between the blood vessel 10 and the retinal region 8 where the luminance value change amount, that is, the edge strength is high, as the boundary between the optic disc region 6 by mistake. This process utilizes the fact that the pixel value of the blood vessel 10 is lower in luminance than the pixel value of the surrounding retinal region 8, and a morphology in which pixels having a pixel value lower than the surrounding pixel value are filled with the surrounding pixel value. It can be executed by applying processing based on computation.

[0030] The following processing performed by the region identification unit to extract the contour of the optic disc region 6 from the stereo images 2 and 4 in which the pixel values of the blood vessel 10 are replaced with the pixel values of the retinal region 8 is roughly described. This is a process for defining an initial contour composed of control points on a plurality of images in order to determine the initial value of the contour of the optic nerve head region 6. Here, the control points constituting the initial contour can be determined by the following processing. That is, a histogram of the pixel values of the stereo images 2 and 4 is created, and threshold processing is performed on the stereo images 2 and 4 with a predetermined luminance value as a threshold value. The center point of the region corresponding to the optic disc region 6 is determined, and the center point force is determined so as to be arranged at equal intervals on a circle at a predetermined distance.

[0031] When the initial contour of the optic disc region 6 is defined by the first control point, the region identification means uses the values of the edge strength around the control point, the distance between the control points, and the slope between the control points as terms. Each term is weighted in consideration of the characteristics of the edge strength of the stereo image, and a conditional expression is applied to perform an operation to find the optimal position of the control point that minimizes the value of the conditional expression. . The position of the control point converges while moving while iterating repeatedly, and the final control point arrangement becomes the contour extraction result. Applying the contour of the optic disc region 6 determined in this way, a stereo image of the fundus showing the optic disc region 6 in white and the retinal region 8 other than the optic disc region 6 and the blood vessel 10 in black is shown in Fig. 4. Show.

[0032] Next, in step s6, the image analysis apparatus analyzes the pixel data of the retinal regions 8 and blood vessels 10 of the stereo images 2 and 4 using the image correction means, Perform alignment. As is clear from the comparison between the left image 2 and the right image 4 in FIG. 2, the stereo image is taken so that the position and shape of the optic disc region 6 differ greatly from image to image. Thus, the retinal region 8 and the blood vessel 10 on the retina are photographed at substantially the same position. The image correction means of the image analysis apparatus can advance the alignment correction with very high accuracy by analyzing the pixel data of the retinal region 8 and the blood vessel 10.

[0033] The image correcting means translates and superimposes the right image 4 on the left image 2 in a vertical direction within a range of dozens of pixels, and overlaps between the two images in the retinal region 8 and the blood vessel 10. Calculate cross-correlation features of pixel values. Similarly, the horizontal correlation is performed by moving the pixels one by one within a range of dozens of pixels and superimposing them, and the cross-correlation features of the pixel values between the two images in the retinal region 8 and the blood vessel 10 are calculated. Here, the cross-correlation feature is obtained by calculating a cross-correlation function using the RGB value, color information such as luminance, saturation, and brightness calculated based on the RGB value, or edge strength. When the cross-correlation feature value is higher than the predetermined reference value by V, it can be considered that the correction of the alignment of the two stereo images has been achieved. The image correction means translates the movement amount in the horizontal direction and the vertical direction when the alignment correction is achieved as a correction value.

[0034] The image correction means evaluates the cross-correlation feature obtained by the parallel movement of the image, and determines that the right image 4 of the stereo image is predetermined when it is determined that more preferable alignment is necessary. It is also possible to calculate the cross-correlation feature quantity with the left image 2 by performing parallel movement after enlarging or reducing in the horizontal direction at a magnification of. This is because it is sometimes possible to obtain a high value of the cross-correlation feature value by enlarging or reducing one image in the horizontal direction. The enlargement / reduction ratio of the right image 4 can indicate the size of the optic disc region 6 identified by the region identifying means. When enlarging the right image 4 in the horizontal direction, the image correcting means inserts a new pixel column at a predetermined interval between the horizontal columns of pixels constituting the image, and the inserted new pixel. Next, the existing existing pixel values are defined by linear interpolation. When the right image 4 is reduced, the horizontal pixel rows are deleted at predetermined intervals.

As described above, the image correction unit of the present embodiment uses the pixel values of the retinal region 8 and the blood vessel 10 of the stereo images 2 and 4 to calculate the cross-correlation feature amount of the pixel values between the two images. Calculate and perform image alignment. The range of the retinal region 8 and blood vessel 10 in the stereo images 2 and 4 is about one-half to about two-thirds. Compared to the case, the analysis proceeds very efficiently. Figure 5 shows the right image 16 with the alignment corrected.

[0036] The cross-correlation feature amount calculated by the cross-correlation function used in the present embodiment has a value even if the luminance is linearly converted between the two stereo images 2 and 4. Since there is no change, it is possible to correct the alignment even for images with different contrasts due to differences in lighting conditions.

[0037] Next, in step s8, the image analysis apparatus analyzes the pixel data of each of the optic disc regions 6 of the stereo images 2 and 16 that have been aligned by the corresponding point determination means, and the optic disc region 6 The corresponding point which is photographing the same part in is extracted and determined. The corresponding point determination means sets a reference point 18 over the entire optic disc region 6 of the left image 2 and searches for a point corresponding to the reference point 18 in the right image 4. In Fig. 6 (a), An example of the reference point set in the optic disc area 6 is shown.

[0038] In order to search for a point corresponding to the reference point 18, the corresponding point determination means sets a rectangular region having a side of several tens of pixels as the region of interest 19 around the reference point 18 in the left image 2. To do. Next, as shown in FIG. 6B, a corresponding region 20 having the same size is set in the vicinity of the region of interest 19 in the left image 2 in the optic disc region 6 in the right image 16. Then, the similarity of the pixel value pattern between the corresponding region 20 and the pixel value of the region of interest 19 is evaluated. If the pattern value of the pixel value of the corresponding region 20 that was initially set is not sufficiently similar to the pattern of the pixel value of the region of interest 19, the corresponding region 20 is moved on the image 16 and the similarity is evaluated again. . Then, the corresponding region 20 having a pattern of pixel values sufficiently similar to the region of interest 19 is searched, and the center of the corresponding region 20 is determined as the corresponding point 21. Corresponding point determination means can calculate a correlation feature between the pixel value of the region of interest 19 and the pixel value of the corresponding region 20 by using a least square matching method, and can evaluate the similarity between stereo images. Even if there is a linear change in color tone, reasonable matching is possible.

The corresponding point determining means moves the corresponding region 20 on the image 16 in step slO, and evaluates the cross-correlation feature quantity of the pixel values of the region of interest 19 and the corresponding region 20. However, if the corresponding region 20 having a sufficiently large cross-correlation feature amount is not obtained even by such movement, the process proceeds to step sl2, and the region of the optic nerve head 6 in the right image 16 is further corrected. Yeah. Corresponding point determination means searches the corresponding region 20 having a pixel data pattern similar to the region of interest 19 again after performing the lateral enlargement or reduction of the optic disc 6 of the right image 16. In the image of the optic nerve head region 6, lateral distortion may occur. In order to cope with such distortion, the corresponding point determination means can determine the corresponding point with higher accuracy by performing lateral enlargement or reduction of the optic nerve head region 6 of the right image 16. When the optic disc region 6 is expanded in the horizontal direction, new pixel columns are inserted at predetermined intervals between horizontal columns of pixels constituting the image, and the inserted new pixel columns are inserted. On the other hand, adjacent existing pixel values are defined by linear interpolation. When the optic nerve head area 6 is reduced, horizontal pixel rows are deleted at predetermined intervals.

In this way, the corresponding point determination unit corrects the optic nerve head region 6 of the right image 16 as necessary, and the region of interest 19 of the reference point 18 set in the left image 2 from the right image 16. A corresponding region 20 having a sufficiently high cross-correlation feature quantity is searched to determine a corresponding point 21 (step s14). Corresponding point determination means determines and stores a set of the reference point 18 of the left image 2 and the corresponding point 21 of the right image 16 over the entire optic disc region 6.

[0041] Next, in step sl6, the image analysis apparatus calculates depth information of the optic nerve head from position information in each image of the stored reference point / corresponding point set using depth information determining means. For the calculation of depth information, a measurement method that applies the principle of triangulation is most commonly applied.

FIG. 7 shows an output example of a three-dimensional image of the optic nerve head region 6 constructed by the image analysis device based on the depth information calculated by the depth information determining means. In the 3D image in Figure 7, what? The image of the left image 2 is applied to the images of the retinal region 8 and blood vessels 10 around the optic disc region 6. In the retinal region 8 and the blood vessel 10, the corresponding points are extracted and the depth information is calculated! Therefore, the fundus image including the exact shape of the optic disc region 6 is output very quickly. In addition, in the display of the retinal region 8, since the corresponding points have not been accurately performed in the past, a structure with large unevenness is displayed even on a smooth flat surface that does not actually have unevenness. However, it is possible to display one image efficiently and smoothly.

[0043] As described above, according to the image analysis apparatus of the present embodiment, the stereo image 2, in which the fundus is photographed, By analyzing the pixel data of 4, it is possible to obtain 3D information that can reproduce the 3D shape of the papillary optic nerve region 6 accurately, efficiently, and quickly. In addition, the image analysis apparatus according to the present embodiment accurately corrects the three-dimensional shape of the optic disc region 6 even if the stereo images 2 and 4 to be analyzed have pixel value changes due to illumination conditions and the like. It can be analyzed.

[0044] Furthermore, when performing stereo image alignment, a sufficient level of cross-correlation features are obtained by performing rotation or deformation of the entire image in addition to parallel translation and enlargement / reduction of the image. You can also It is also possible to obtain more accurate and reliable information indicating the three-dimensional shape using three or more stereo images.

[0045] Although one embodiment of the present invention has been described in detail above in the embodiment, this is merely an example and does not limit the scope of the claims. The technology described in the claims includes various modifications and changes of the specific examples illustrated above. For example, the image analysis apparatus in this embodiment can be replaced with an image analysis program that can cause a computer to execute the same process. Further, each means of the image analysis apparatus of the present embodiment can be configured as an external apparatus that is modularized and connected to a computer. Furthermore, the geometrical method of image analysis employed by the area identification unit, image correction unit, corresponding point determination unit, and depth information determination unit of the image analysis apparatus is the characteristics of the image content and the measurement object. It can be selected and changed as appropriate.

Brief Description of Drawings

FIG. 1 is a diagram showing a basic imaging position when a stereo image of the fundus of a subject is captured. FIG. 2 is a diagram showing an example of stereo images 2 and 4 to be analyzed.

FIG. 3 is a flowchart showing the contents of image processing by the image analysis apparatus of one embodiment of the present invention.

FIG. 4 is a stereo image diagram of the fundus oculi where the optic disc region 6 is shown in white and the retinal region 8 and blood vessels 10 other than the optic disc region 6 are shown in black by the region identification process.

FIG. 5 is a diagram showing a right image 16 that has been subjected to alignment correction. FIG. 6 shows a left image 2 in which a region of interest 19 is arranged and a right image 4 in which a corresponding region 20 is arranged.

FIG. 7 is a diagram showing an output example of a three-dimensional image of the optic nerve head region 6 constructed by the image analysis apparatus.

Explanation of symbols

2 Left image 4, 16 Right image 6 Optic nerve head region 8 Retina region 10 Blood vessel 12 Papillary margin 14 Depression recess 18 Reference point 19 Region of interest 20 Corresponding region 21 Corresponding point 22, 24 Imaging position 26, 28 image

Claims

The scope of the claims

[1] An image analyzing apparatus for analyzing a three-dimensional shape of a optic nerve head from a stereo image obtained by photographing a fundus of a subject, an image input means for inputting pixel data of the stereo image, and each stereo image The region discriminating means for discriminating between the optic disc region and the region other than the optic disc, and the pixel data of the region other than the optic disc in the stereo image are analyzed to align the entire image including the optic disc region. The image correction means to be performed and the pixel data of the region of the optic nerve head of the stereo image aligned by the image correction means are analyzed, and corresponding points to be captured are determined by photographing the same portion in the optic nerve head. Corresponding point determining means, and depth information determining means for calculating depth information of the optic nerve head from position information in each image of the corresponding points determined by the corresponding point determining means. In the image analysis apparatus according to claim Rukoto.

[2] The image correction means evaluates the total difference between the pixel values at the same position between the stereo images, and determines the vertical movement amount, the horizontal movement amount, and the horizontal scaling ratio between the images. 2. The image analysis apparatus according to claim 1, wherein the position of the image is corrected by obtaining.

[3] Corresponding point determination means sets a reference point in the area of the optic nerve head of the reference image, sets a region of interest in a specific range around this reference point, and sets the optic nerve head of another image. A region having a pixel data array most similar to the pixel data of the region of interest is searched from the region, and an array of pixel data sufficiently similar to the region of interest is not obtained in the region of the optic nerve head of another image. In this case, the region of the optic nerve head in the other image is expanded or reduced in the horizontal direction to search for a region having an array of pixel data most similar to the pixel data of the region of interest, and the result of the search The image analysis apparatus according to claim 1 or 2, wherein a corresponding point between the image used as a reference and another image is determined based on the above.

[4] An image analysis program for analyzing the three-dimensional shape of the optic nerve head from pixel data of a stereo image obtained by photographing the fundus of the subject, wherein the area of the optic nerve head and the area other than the optic nerve head for each of the stereo images Are identified, and the pixel data of the region other than the optic nerve head of the stereo image are analyzed. Align the entire image including the optic disc area! ヽ, the optic disc area of the aligned stereo image The processing of calculating the depth information of the optic disc from the position information in each image of the corresponding point is determined by analyzing An image analysis program to be executed.