TWI629969B - Eye detection system and method - Google Patents

Abstract

An eye detection system includes an image capture device and an image determination device. The image capturing device is configured to capture a measured image to obtain a measured image, and the image determining device is coupled to the image capturing device. The image determining device includes a measurement image processing unit, a database, and an image evaluation unit. The measurement image processing unit is configured to analyze the measurement image to obtain at least one attribute value corresponding to the measurement image; the database stores a plurality of eye images and quality status information corresponding to the plurality of eye images; The image evaluation unit estimates the image quality of the measured image by using the at least one attribute value of the measured image according to the plurality of eye images in the database and the quality status information corresponding to the plurality of eye images status.

Description

Eye detection system and method

The present disclosure relates to an eye detection system and method.

Due to the professional division of labor between the physician and the optometrist, the eye examination is often diagnosed by the optometrist (Optometrist) who takes the picture and then transmits the fundus map to the back end. If the doctor can't immediately view the fundus map, sometimes the photo that the medical examiner sends to the doctor will occur. Because the doctor can't read it, the patient must be asked to re-take the sample, which causes a lot of inconvenience.

With the demand for mobile medical materials, mobile fundus cameras have been developed. In order to facilitate the convenience, the camera is usually randomly set to have a small screen size, and the hand-held medical material is easy to shake, so it is often difficult for the photographer to directly judge the quality of the photograph from the screen. If the interval between the front and back ends is too long, it is not easy to get proper feedback immediately, so an indicator is needed to remind the photographer of the image quality of the current shooting.

One embodiment of the present disclosure provides an eye detection system that includes an image capture device and an image determination device. The image capturing device is configured to capture a measured image to obtain a measured image, and the image determining device is coupled to the image capturing device. The image determining device includes a measurement image processing unit, a database, and an image evaluation unit. among them, The measurement image processing unit is configured to analyze the measurement image to obtain at least one attribute value corresponding to the measurement image; the database stores a plurality of eye images and quality status information corresponding to the plurality of eye images; the image evaluation The unit evaluates the image quality status of the measurement image by using the plurality of eye images in the database and the quality status information corresponding to the plurality of eye images, and the at least one attribute value of the measurement image.

One embodiment of the present disclosure proposes an eye detection method. The method includes: capturing an image by one of the examined eyes by an image capturing device; and receiving and analyzing the measured image by an image determining device to evaluate an image quality state of the measured image. The step of analyzing the measurement image includes: analyzing the measurement image to obtain at least one attribute value corresponding to the measurement image; and the plurality of eye images provided according to a database and the quality corresponding to the plurality of eye images Status information, the image quality status of the measurement image is evaluated by the at least one attribute value of the measurement image.

10‧‧‧Eye detection system

12‧‧‧Image capture device

14‧‧‧Image Judging Device

142‧‧‧Measurement image processing unit

144‧‧‧Database

146‧‧‧Image Evaluation Unit

16‧‧‧Output device

FI‧‧‧Measurement image

Ai‧‧‧ attribute value

A1‧‧‧ sharpness

A2‧‧‧Brightness

A4‧‧‧Contrast

A4‧‧‧Saturation

QI‧‧‧Image quality status

300A, 300B, 400A, 400B, 400C, 500A, 500B, 500C, 600A, 600B, 600C‧‧‧ distribution curves

T1, T2‧‧‧ quality acceptance values

T14, Tr4, T16, Tr6‧‧

R1, R2‧‧‧ pattern box selection

P(μ)‧‧‧Quality indicator probability model

‧‧‧‧Quality index

P ₁ (μ ₁ )‧‧‧ sharpness probability model

P ₂ (μ ₂ )‧‧‧Brightness probability model

P ₃ (μ ₃ )‧‧‧saturation probability model

Probability model of P ₄ (μ ₄ )‧‧‧ contrast

μ ₁ ‧‧‧ sharpness parameter

μ ₂ ‧‧‧Brightness parameters

μ ₃ ‧‧‧Saturation parameters

μ ₄ ‧‧‧Contrast parameters

14’‧‧‧Physician Dispatch System

144’‧‧‧Physician Database

D1~Dn‧‧‧Physician

Db1~dbn‧‧‧Eye image data group

ES1~ESn‧‧‧Quality Status Information

1 is a block diagram showing an eye detecting system according to an embodiment of the present disclosure.

2 is a flow chart showing the establishment of a quality indicator probability model according to an embodiment of the present disclosure.

FIG. 3A is a statistical distribution diagram of sharpness of an experimental example.

FIG. 3B is a statistical distribution diagram of brightness of an experimental example.

FIG. 3C is a statistical distribution diagram of saturation of an experimental example.

FIG. 3D is a statistical distribution diagram of contrast of an experimental example of the present disclosure.

FIG. 3E is a schematic diagram of a quality index probability model of an experimental example.

4 is a flow chart showing an eye detecting method according to an embodiment of the present disclosure.

Figure 5 shows a schematic diagram of a physician dispensing system in accordance with an embodiment of the present disclosure.

1 is a schematic diagram showing an eye detection system 10 in accordance with an embodiment of the present disclosure. Referring to FIG. 1 , the eye detection system 10 includes an image capturing device 12 , an image determining device 14 , and an output device 16 . In this embodiment, the image determining apparatus 14 includes a measurement image processing unit 142, a database 144, and an image evaluation unit 146, and each unit in the image determining apparatus 14 (for example, the measurement image processing unit 142, the database 144, and The image evaluation unit 146) can communicate with each other.

In the embodiment, the image capturing device 12 may be a camera, a camera, or the like, but the disclosure is not limited thereto. In an embodiment, the image capturing device 12 can also be an image information transmission interface device for receiving image information output from any image providing end.

In the present embodiment, the image determining device 14 may be a device implemented by at least one integrated circuit. In another embodiment, the image determination device 14 can also be implemented by a program executed by the processor. The database 144 may be a memory device, and the database 144 may be disposed in a hardware device including the image determining device 14, but the disclosure is not limited thereto. In one embodiment, the database 144 may also be a memory device that is wired or wirelessly communicated with the image determining device 14, such as an external hard disk or a cloud hard disk server.

Referring to the embodiment of FIG. 1 , the image capturing device 12 is configured to take a picture of the eye to be inspected to obtain a measurement image FI corresponding to the eye to be inspected, and the measurement image FI is, for example, a fundus image. The image capturing device 12 can be a slit lamp or other visual inspection device capable of taking images.

In this embodiment, the image determining device 14 is coupled to the image capturing device 12 to enable The measurement image processing unit 142 receives the measurement image FI and analyzes the measurement image FI to obtain an attribute value PI corresponding to the measurement image FI. The image evaluation unit 146 receives the attribute value A of the measurement image FI, and based on the data stored in the database 144, and evaluates the image quality state QI of the measurement image FI by the attribute value PI, the quality status information may be For example, information such as good, poor, normal, abnormal, qualified, unqualified, too bright or too dark.

In the embodiment, the output device 16 is coupled to the image determining device 14 for outputting the quality status information QI of the measurement image FI, but the disclosure is not limited thereto. In an embodiment, the output device 16 may be an external output device coupled to the eye detection system for outputting the quality status information QI of the measurement image FI. The output device 16 can output the quality status information QI of the measurement image FI in a manner such as a picture, a sound, a character, or a light color, but the disclosure is not limited thereto. The output device 16 can be, for example, a display panel, a horn, a signal light, or any device that can output display information. In this embodiment, the output device 16 is, for example, a display panel. When the quality status information QI of the measurement image FI is evaluated as being good, the display panel (output device 16) displays a green light number in the display screen; When the quality status information QI of the measurement image FI is evaluated as poor, the display panel (output device 16) displays a green light number on the display screen, and a person (eg, a photographer) operating the eye detection system 10 can transmit the output device. The 16-output signal determines whether the eye to be inspected needs to be re-photographed.

In this embodiment, the image evaluation unit 146 may further store the measurement image FI and the corresponding quality status information QI in the data repository 144.

In this embodiment, the data stored in the database 144 includes a plurality of eye images and quality status information of the eye images, and the quality status information may be, for example, good, poor, normal, abnormal, qualified, unqualified, Information such as too bright or too dark. In this embodiment, the quality status information may be the interpretation result information of a professional such as a doctor or an optometrist, but the disclosure is not limited. herein. In an embodiment, the quality status information may also be the interpretation result information of the smart machine system or the quality status information fed back by the image evaluation unit 146.

In an embodiment, the image evaluation unit 146 can also be an image analysis unit, and the image analysis unit can receive the data stored in the database 144 and analyze the data according to the statistical probability model to establish a quality index probability model. . In an embodiment, the image evaluation unit 146 can receive the attribute value PI of the measurement image FI from the measurement image processing unit 142, and evaluate the measurement image FI by the attribute value PI according to the established quality index probability model. Image quality status information QI.

FIG. 2 is a schematic flow chart showing the establishment of a quality indicator probability model according to an embodiment of the present disclosure. The image analysis unit (eg, the image evaluation unit 146) obtains a plurality of eye images and quality state information of the eye images from the database 144, wherein the plurality of eye images are a plurality of fundus images captured by the eye detection system 10, The quality status information of the eye images is information corresponding to the results of the interpretation of the fundus images of the various fundus images of the plurality of fundus images. In step S601, at least one attribute value Ai of the fundus image is calculated by the image evaluation unit 146, where i is an integer greater than 0, and the attribute value may include sharpness A1, brightness A2, contrast A4 or saturation A4, and the like. In an embodiment, before calculating the attribute value of the fundus image, the image analyzing unit (for example, the image evaluating unit 146) may further detect the fundus image to select an area of interest from the fundus image (Region of In the image of Interest, ROI), the range of the region of interest is, for example, a retinal image region, and in step S601, the image evaluation unit 146 may calculate only at least one attribute of the image in the region of interest circled from the fundus image. value.

In step S602, the image evaluation unit 146 calculates the at least one attribute value corresponding to the fundus image and the quality of the eye image corresponding to the fundus image calculated according to step S601. Status information to obtain a statistical distribution of each attribute value corresponding to at least one attribute value and each quality status information.

Please refer to the corresponding attribute values of the experimental examples of FIG. 3A to FIG. 3D as the statistical distribution maps of sharpness, brightness, saturation and contrast.

Fig. 3A shows a statistical distribution of sharpness of an experimental example in which the horizontal axis is sharpness and the vertical axis is number of images. FIG. 3A is a view showing the distribution (sharpness) and sharpness distribution of the interpretation results of the plurality of fundus images, and the distribution curve 300A is the distribution of the sharpness of the fundus images which is better as the information of the interpretation results. The curve 300B is a distribution of the sharpness of the fundus images in which the information of the interpretation result is poor. In the present experimental example, the numerical value represented by the horizontal axis is a value which is normalized and distributed between 0 and 100, but the disclosure is not limited thereto.

Fig. 3B shows a statistical distribution of brightness of an experimental example in which the horizontal axis is brightness and the vertical axis is the number of images. FIG. 3B is a view showing the distribution of information (normal, too dark or too bright) and brightness (Brightness) corresponding to the reading results of a plurality of fundus images, and the distribution curve 400A is the brightness of the fundus images in which the information of the interpretation result is normal. The distribution, the distribution curve 400B is the distribution of the brightness of the fundus images that are too dark for the interpretation result, and the distribution curve 4C is the distribution of the brightness of the fundus images that are too bright for the interpretation result.

Fig. 3C shows a statistical distribution of saturation of an experimental example in which the horizontal axis is saturation and the vertical axis is the number of images. FIG. 3C is a view showing distribution of information (good or bad) and saturation (Saturation) corresponding to the interpretation results of the plurality of fundus images, and the distribution curve 500A is a distribution of the saturation of the fundus images in which the information of the interpretation result is normal. The curve 500B is the distribution of the saturation of the fundus images that are too dark for the interpretation result, and the distribution curve 500C is the interpretation result. The information is the distribution of the saturation of the fundus images that are too bright.

Fig. 3D shows a statistical distribution of the contrast of the experimental example, in which the horizontal axis is the contrast and the vertical axis is the number of images. FIG. 3D is a view showing the distribution of the information (good or bad) and contrast (Contrast) corresponding to the interpretation results of the plurality of fundus images, and the distribution curve 600A is the distribution of the contrast of the fundus images in which the information of the interpretation result is normal, and the distribution curve 600B The information for the interpretation result is the distribution of the contrast of the fundus images that are too low, and the distribution curve 600C is the distribution of the contrast of the fundus images that are too large for the interpretation result.

In step S603, the image evaluation unit 146 analyzes the statistical distribution of the corresponding attribute values and quality status information obtained in step S602 to set an Acceptable Quality Level (AQL) corresponding to the attribute value, wherein the quality The acceptance criteria can be a numerical value or a numerical range.

Please refer to Figure 3A to Figure 3D of the experimental example. Referring to the statistical distribution of the sharpness of FIG. 3A, the image evaluation unit 146 analyzes the distribution curve 300A and the distribution curve 300B to set the quality acceptance value T1 corresponding to the attribute value as the sharpness, and the right side of the quality acceptance value T1 (eg, the corresponding quality). The right side of the dotted line of the allowable value T1 is the range of acceptable sharpness, and the left side of the quality acceptance value T1 (as indicated by the left side of the dotted line corresponding to the quality acceptance value T1) is an unacceptable range of sharpness. The limit value of the distribution curves 300A and 300B in the color space can be distinguished by the quality acceptance value T1. In this experimental example, the quality acceptance value T1 is, for example, a sharpness of 15.

Referring to the statistical distribution of the brightness of FIG. 3B, the image evaluation unit 146 analyzes the distribution curve 400A, the distribution curve 400B, and the distribution curve 400C to set R1 as the quality acceptance range of the corresponding attribute value for brightness, and the range of the pattern frame selection area R1. In order to distinguish the range between normal, too dark, and too bright, wherein the limit Tl4 is distinguishable from the distribution curve 400A and the distribution curve 400B In the limit value of brightness, the limit Tr4 is a boundary value between the brightness distribution 400A and the distribution curve 400C, and the frame selection area R1 is a frame selection area from the limit T14 to the limit Tr4. In this experimental example, the pattern frame selection area is selected. The range of R1 is, for example, a range of brightness ranging from 15 to 32.

Referring to the statistical distribution of saturation of FIG. 3C, the image evaluation unit 146 analyzes the distribution curve 500A and the distribution curve 500B and/or the distribution curve 500A and the distribution curve 500C to set the quality acceptance value T2 corresponding to the attribute value as the saturation value. The right side of the allowable value T2 (as indicated by the right side of the dotted line corresponding to the quality acceptance value T2) is the range of acceptable saturation, and the left side of the quality acceptance value T2 (such as the left side of the dotted line corresponding to the quality acceptance value T2) The indication is an unacceptable range of saturation, wherein the quality acceptance value T2 is a threshold value at which the distribution curves 500A and 500B can be resolved. In the present experimental example, the quality acceptance value T2 is, for example, a sharpness of 75.

Referring to the statistical distribution of the contrast of FIG. 3D, the image evaluation unit 146 analyzes the distribution curve 600A, the distribution curve 600B, and the distribution curve 600C to set R2 as the quality acceptance range of the corresponding attribute value, and the range of the pattern frame selection area R2 is The range between normal, too dark, and too bright is distinguished, wherein the limit T16 is a boundary value between the distribution curve 600A and the distribution curve 600B, and the boundary Tr6 is a boundary value between the distribution curve 600A and the distribution curve 600C. The pattern frame selection area R2 is a frame selection area of the boundary T16 to the boundary Tr6. In the present experimental example, the range of the pattern frame selection area R2 is, for example, a range of contrast of 10 to 20.

In step S604, the image evaluation unit 146 establishes a quality index probability model P(μ) according to the curve distribution corresponding to each attribute value and the quality acceptance criteria of each attribute value, where μ is a quality index parameter (Quality Index). In this embodiment, the probability model corresponding to each attribute value can be obtained by the distribution curve of the experimental example of FIG. 3A to FIG. 3D and the quality acceptance criteria of the attribute values, for example, P ₁ (μ ₁ ) is the corresponding attribute. The probability model with sharpness value, P ₂ (μ ₂ ) is the probability model corresponding to the attribute value of brightness, P ₃ (μ ₃ ) is the probability model corresponding to the attribute value of saturation, and P ₄ (μ ₄ ) is corresponding. The attribute value is a probability model of contrast, where μ ₁ , μ ₂ , μ _{3 ,} and μ ₄ are a sharpness, a brightness, a saturation, and a contrast, respectively. Referring to FIG. 3E, the quality index probability model P(μ) may be established by at least one probability model of the probability models P ₁ -P ₄ . In this embodiment, the quality index probability model P(μ) may be represented as P(μ). =P ₁ ×P ₂ ×P ₃ ×P ₄ , but the disclosure is not limited thereto.

FIG. 4 is a flow chart showing an eye detecting method according to an embodiment of the present disclosure. In step S101, the image capturing device 12 captures the eye to be inspected to obtain a fundus image FI corresponding to the eye to be examined.

In step S102, the measurement image processing unit 142 receives and analyzes the fundus image FI of the subject's eye to obtain at least one attribute value corresponding to the fundus image FI, wherein the attribute value includes sharpness, brightness, saturation, or contrast. .

In an embodiment, the measurement image processing unit 142 may first detect the fundus image FI to circle an image in a Region of Interest (ROI) from the fundus image, the range of the region of interest. For example, in the retinal image region, in step S102, the measurement image processing unit 142 may calculate only at least one attribute value PI of the image in the region of interest corresponding to the fundus image FI.

In step S103, the image evaluation unit 146 receives the attribute value PI corresponding to the fundus image FI, and determines the quality index probability model P(μ) established according to the embodiment of FIG. 2 and the quality acceptance standard corresponding to the attribute value. Quality status information QI of fundus image FI.

In step S104, the output device 16 outputs the quality status information QI of the measurement image FI.

Figure 5 shows a schematic diagram of a physician dispensing system 14' in accordance with an embodiment of the present disclosure. in In practice, because each doctor's eye response performance will be different, for example, some doctors are more sensitive to brighter images, some doctors still have good analytical ability for darker or less obvious images, or some physicians have Reddish images are more sensitive. The physician assignment system 14 of the embodiment of Fig. 5 is similar to the image determination device 14 of the embodiment of Fig. 1, including a measurement image processing unit 142, a database 144, and an image evaluation unit 146. The database 144 can store the physician identification information corresponding to the plurality of eye image data (for example, the name of the doctor, etc.), so the data stored in the database 144 can be classified according to the physician identification information to form a physician database. 144'. The image evaluation unit 146 may analyze at least one attribute value of the plurality of eye images in the physician database according to the physician identification information, and according to the statistical probability model and the physician identification information, by using the plurality of eye images The attribute values and the quality status information establish a plurality of quality indicator probability models corresponding to the identification information of the physicians.

Referring to the embodiment of FIG. 5, the physician database 144' stores a plurality of eye image data groups db1 to dbn corresponding to the doctors D1 to Dn, respectively, where n is an integer not less than one. In the present embodiment, the quality index probability models P1(μ) to Pn(μ) corresponding to the physicians D1 to Dn, respectively, can be obtained according to the foregoing steps S601 to S604. When the image capturing device 12 of the present embodiment is configured to capture the measured image of the eye to be inspected, the image evaluation unit 146 may perform the foregoing step S103 according to the physicians D1 to Dn, respectively. The quality indicator probability models P1 (μ) to Pn (μ) are obtained to obtain quality state information ES1 to ESn corresponding to the fundus images FI of the physicians D1 to Dn, respectively.

In this embodiment, the physician assignment system 14' may further include a dispatch module 148 for assigning the fundus image FI to at least the physicians D1 to Dn according to the quality status information ES1 to ESn of the physicians D1 to Dn, respectively. A physician, but the disclosure is not limited to this. In an embodiment The dispatch module 148 can output a dispatch recommendation information to the output device 16 according to the quality status information ES1 to ESn of the physicians D1 to Dn, respectively.

The disclosure of the present invention is disclosed in the foregoing embodiments, and is not intended to limit the disclosure. Any skilled person can make some modifications and refinements without departing from the spirit and scope of the disclosure. The scope of protection shall be subject to the definition of the scope of the patent application attached to this specification.

Claims (14)

An eye detection system includes: an image capturing device for capturing a test eye to obtain a measurement image; and an image determination device coupled to the image capture device, the image determination device having: a measurement An image processing unit for analyzing the measurement image to obtain at least one attribute value corresponding to the measurement image; a database for storing a plurality of eye images and quality status information corresponding to the plurality of eye images; and an image evaluation a unit, according to the plurality of eye images in the database and the quality status information corresponding to the plurality of eye images, the plurality of eye images being a plurality of fundus images captured by the image capturing device, the eyes The image quality status information is corresponding to each fundus image of the plurality of fundus images, and the image quality state of the measurement image is evaluated by the at least one attribute value of the measurement image; the image evaluation unit analyzes the image quality in the database At least one attribute value of the plurality of eye images, and according to the statistical probability model, the at least one attribute value of the plurality of eye images and the plurality of Quality status information to get for each attribute value should be at least one attribute value of the state and various quality of statistical information distribution, establish a quality indicator probability models. The eye detection system of claim 1, comprising: An output device is configured to output status information of the image quality status of the measurement image. The eye detection system of claim 1, wherein the at least one attribute value corresponding to the measurement image comprises sharpness, brightness, saturation, and contrast. The eye detection system of claim 1, wherein the image evaluation unit detects the fundus image to circle an image in the region of interest from the fundus image. The eye detection system of claim 4, wherein the at least one attribute value corresponding to the plurality of eye images comprises sharpness, brightness, saturation, and contrast. The eye detection system of claim 4, wherein the image evaluation unit evaluates an image quality state of the measurement image from the at least one attribute value of the measurement image according to the quality indicator probability model. The eye detection system of claim 1, wherein the image evaluation unit stores the measurement image and the quality status information corresponding to the measurement image in the database. The eye detection system of claim 1, wherein the database stores a plurality of physician identification information of the quality status information of the plurality of eye images. The eye detection system of claim 8, wherein the plurality of eye images stored in the database and the quality status information are classified according to the physician identification information to form a physician database. The eye detection system of claim 9, wherein the image evaluation unit analyzes at least one attribute value of the plurality of eye images in the physician database according to the physician identification information, and according to the statistical probability model and the The physician identification information is used to establish a plurality of quality indicator probability models corresponding to the physician identification information by using the attribute values of the plurality of eye images and the quality status information. The eye detection system of claim 10, wherein the image evaluation unit evaluates the at least one attribute value of the measurement image according to the plurality of quality indicator probability models respectively corresponding to the physician identification information Multiple image quality states of the measurement image corresponding to some physician identification information. The eye detection system of claim 11, comprising: a dispatching unit, configured to assign the measured image to the pair according to the image quality states of the measured images respectively corresponding to the physician identification information At least one of the physicians should be identified by the physician. An eye detecting method comprising: capturing an image of one of the examined eyes by an image capturing device; Receiving and analyzing the measurement image by an image determination device to evaluate an image quality state of the measurement image, wherein the step of analyzing the measurement image comprises: analyzing the measurement image to obtain at least the corresponding measurement image An attribute value; and a plurality of eye images provided according to a database and quality status information corresponding to the plurality of eye images, the plurality of eye images being a plurality of fundus images captured by the image capturing device, the eyes The quality status information of the image is the fundus image corresponding to the plurality of fundus images, and the image quality state of the measured image is evaluated by the at least one attribute value of the measured image; and the plurality of eye images in the database are analyzed. At least one attribute value; according to the statistical probability model, the attribute values of the plurality of eye images and the quality status information are used to obtain a statistical distribution of each attribute value corresponding to at least one attribute value and each quality status information Establishing a quality indicator probability model; and evaluating the quantity from the at least one attribute value of the measurement image according to the quality indicator probability model Image quality state image. The method for detecting an eye according to claim 13 , wherein the at least one attribute value of the image is measured according to the plurality of eye images provided by the database and quality state information corresponding to the plurality of eye images The step of evaluating the image quality state of the measurement image includes: first detecting the fundus image to circle an image in the region of interest from the fundus image.