WO2021256595A1 - Vision testing device, method, and program stored in computer-readable storage medium - Google Patents

Abstract

An embodiment of the present invention relates to a vision testing device, method, and program stored in a computer-readable storage medium, and a technical problem to be solved is to provide a vision testing device, method, and program stored in a computer-readable storage medium, capable of calculating vision information (for example, a refractive index of an eye) from an eye-photographed image via a user device. To this end, the present invention provides a vision testing device comprising: a user device for capturing and transmitting a pupil image; and a server which has a super-resolution generative adversarial network (SRGAN) module for receiving a pupil image from the user device and acquiring a super-resolved (SR) image, an object detection module for receiving an SR image and acquiring result box data, and a vision information calculation function module for receiving result box data to calculate vision information, wherein the server transmits the calculated vision information to the user device.

Description

Vision test apparatus, method, and program stored in a computer-readable storage medium

An embodiment of the present invention relates to an eye examination apparatus, a method, and a program stored in a computer-readable storage medium.

Recently, the spread of user devices such as smart phones and tablet PCs and IT infrastructure are rapidly expanding. In general, the display unit of the user device displays various texts and images, and accordingly, users are exposed to the display of the user device regardless of time and place. Accordingly, the eyesight of modern people is gradually deteriorated.

In general, visual acuity may be measured by a measurer maintaining a certain distance at a hospital, a public health center, an optician's shop, or the like, seeing and responding to a plurality of letters or figures displayed on an optometry table according to an instruction of a subject. This conventional method for measuring acuity has a disadvantage in that the reliability of the measured value for eyesight is low. In addition, in order to know one's own visual acuity, it is inconvenient to visit a place where visual acuity can be measured, that is, a hospital. Accordingly, there is a need in the art for convenient and precise vision measurement.

The above-described information disclosed in the background technology of the present invention is only for improving the understanding of the background of the present invention, and thus may include information that does not constitute the prior art.

An object to be solved according to an embodiment of the present invention is an eye examination apparatus capable of calculating vision information (eg, refraction of the eye) from an eye photographed image through a user device, a method, and a method stored in a computer-readable storage medium to provide the program.

A vision test method according to an embodiment of the present invention comprises the steps of receiving a pupil image; inputting the pupil image into an SRGAN (Super-Resolution Generative Adversarial Network) module; obtaining an SR (Super-Resolved) image from the SRGAN module; inputting the SR image to the object detection module; obtaining result box data from the object detection module; inputting the result box data into an eyesight information calculation function module; and obtaining and transmitting the vision information from the vision information calculation function module.

The pupil image receiving step may be performed by receiving a pupil image from a user device including a lighting unit and a photographing unit, and capable of wired or wireless communication.

The SRGAN module can output the SR image by removing noise from the received pupil image and increasing the resolution using the learned SRGAN algorithm.

The SRGAN module includes a generator network that receives a noise image and generates a fake pupil image, and a discriminator network that receives and predicts an actual pupil image and a fake pupil image. learning can be done.

The SRGAN module uses perceptual similarity instead of adversarial loss and pixel space similarity to train a generator network that outputs the SR-imaged perceptual loss including content loss. You can provide a perceptual loss function.

The object detection module may perform object detection using deep learning to analyze the size and position of the iris, pupil, and pupil reflection patterns, and correct the object box through a post-processing algorithm.

The object detection module may use at least one of a Single Shot MultiBox Detector (SSD), YOLOv3, or EfficientDet.

The transmitting of the vision information may be performed by transmitting the vision information to a user device capable of wired or wireless communication.

The vision information may include eye refraction.

A computer program stored in a computer-readable storage medium for eye examination according to an embodiment of the present invention includes: receiving a pupil image; inputting the pupil image to the SRGAN (Super-Resolution Generative Adversarial Network) module; obtaining an SR (Super-Resolved) image from the SRGAN module; inputting the SR image to the object detection module; obtaining result box data from the object detection module; inputting the result box data into the vision information calculation function module; and obtaining and transmitting the vision information from the vision information calculation function module.

An eyesight test apparatus according to an embodiment of the present invention includes: a user device for capturing and transmitting a pupil image; and an SRGAN (Super-Resolution Generative Adversarial Network) module that receives the pupil image received from the user device and acquires an SR (Super-Resolved) image, and an object detection module that receives the SR image and acquires result box data; , a server having an eyesight information calculation function module that receives the result box data and calculates vision information, and the server may transmit the calculated vision information to the user device.

The user device may include a lighting unit and a photographing unit, and may transmit the photographed pupil image to the server through a wired or wireless communication method.

The SRGAN module can output the SR image by removing noise from the received pupil image and increasing the resolution using the learned SRGAN algorithm.

The SRGAN module includes a generator network that receives a noise image and generates a fake pupil image, and a discriminator network that receives and predicts an actual pupil image and a fake pupil image. learning can be done.

The SRGAN module uses perceptual similarity instead of adversarial loss and pixel space similarity to train a generator network that outputs the SR-imaged perceptual loss including content loss. You can provide a perceptual loss function.

The object detection module may perform object detection using deep learning to analyze the size and position of the iris, pupil, and pupil reflection patterns, and correct the object box through a post-processing algorithm.

The object detection module may use at least one of a Single Shot MultiBox Detector (SSD), YOLOv3, or EfficientDet.

The vision information may include eye refraction.

An embodiment of the present invention provides an apparatus for eye examination, a method, and a program stored in a computer-readable storage medium capable of calculating vision information (eg, refractive index of an eye) from an image taken by an eyeball through a user device.

1 is a schematic diagram showing the configuration of an eye examination apparatus according to an embodiment of the present invention.

FIG. 2 is a block diagram illustrating the configuration of a user device and/or a server in the vision test apparatus according to an embodiment of the present invention.

3 is a block diagram illustrating the configuration of program codes stored in a memory of a user device and/or a server among the vision test apparatuses according to an embodiment of the present invention.

4 is a flowchart illustrating an eyesight examination method according to an embodiment of the present invention.

5 is a flowchart illustrating an image analysis method among the visual acuity testing methods according to an embodiment of the present invention.

6A and 6B are block diagrams illustrating a learning algorithm using a Generative Adversarial Network (GAN) and a Super-Resolution Using a Generative Adversarial Network (SRGAN) in an image analysis method according to an embodiment of the present invention.

7 is a block diagram illustrating a learning algorithm using a Single Shot MultiBox Detector (SSD) and YOLO in an image analysis method according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Examples of the present invention are provided to more completely explain the present invention to those of ordinary skill in the art, and the following examples may be modified in various other forms, and the scope of the present invention is as follows It is not limited to an Example. Rather, these examples are provided so that this disclosure will be more thorough and complete, and will fully convey the spirit of the invention to those skilled in the art.

In addition, in the following drawings, the thickness or size of each layer is exaggerated for convenience and clarity of description, and the same reference numerals in the drawings refer to the same elements. As used herein, the term “and/or” includes any one and all combinations of one or more of those listed items. In addition, in the present specification, "connected" means not only when member A and member B are directly connected, but also when member A and member B are indirectly connected with member C interposed between member A and member B. do.

The terminology used herein is used to describe specific embodiments, not to limit the present invention. As used herein, the singular form may include the plural form unless the context clearly dictates otherwise. Also, as used herein, “comprise, include” and/or “comprising, including” refer to the referenced shapes, numbers, steps, actions, members, elements, and/or groups thereof. It specifies the presence and does not exclude the presence or addition of one or more other shapes, numbers, movements, members, elements and/or groups.

Although the terms first, second, etc. are used herein to describe various members, parts, regions, layers and/or parts, these members, parts, regions, layers, and/or parts are limited by these terms so that they It is self-evident that These terms are used only to distinguish one member, component, region, layer or portion from another region, layer or portion. Accordingly, a first member, component, region, layer or portion discussed below may refer to a second member, component, region, layer or portion without departing from the teachings of the present invention.

In addition, the control unit (controller) and/or other related devices or components according to the present invention may be implemented using any suitable hardware, firmware (eg, application specific semiconductor), software, or a suitable combination of software, firmware and hardware. can For example, various components of a control unit (controller) and/or other related devices or parts according to the present invention may be formed on one integrated circuit chip or on separate integrated circuit chips. In addition, various components of the control unit (controller) may be implemented on a flexible printed circuit film, a tape carrier package, a printed circuit board, or may be formed on the same substrate as the control unit (controller). In addition, various components of the control unit (controller), in one or more computing devices, may be processes or threads executing in one or more processors, which execute computer program instructions to perform various functions mentioned below. and interact with other components. The computer program instructions are stored in a memory that can be executed in a computing device using a standard memory device, such as, for example, a random access memory. The computer program instructions may also be stored in other non-transitory computer readable media such as, for example, a CD-ROM, flash drive, and the like. In addition, those skilled in the art related to the present invention are skilled in the art that functions of various computing devices are combined with each other, integrated into one computing device, or functions of a specific computing device are one or more other computing devices without departing from the exemplary embodiments of the present invention. It should be recognized that they can be distributed among

For example, the control unit (controller) according to the present invention is usually composed of a central processing unit, a mass storage device such as a hard disk or a solid state disk, a volatile memory device, an input device such as a keyboard or mouse, and an output device such as a monitor or printer. It can be run on a commercial computer of

1 is a schematic diagram illustrating a configuration of an eye examination apparatus 100 according to an exemplary embodiment of the present invention.

As shown in FIG. 1 , the vision test apparatus 100 according to an embodiment of the present invention may include a user device 110 , a server 120 , and an Internet network 130 .

The user device 110 may basically photograph a pupil image and transmit the photographed pupil image to the server 120 through the Internet network 130 . Also, the user device 110 may receive vision information (eg, the refractive index of the eye) calculated from the server 120 through the Internet network 130 . In some examples, user device 110 may include a smartphone, tablet, notebook, or desktop computer.

The server 120 basically receives the pupil image from the user device 110 and may calculate vision information by analyzing the pupil image using various neural networks. The server 120 may transmit the calculated visual acuity information back to the user device 110 through the Internet network 130 .

As will be described again below, the server 120 pre-processes using a super resolution technique using Generative Adversarial Networks (GAN) to perform accurate object detection from the received pupil image, and then deep Analyzes the size and position of the iris, pupil, and pupil reflection patterns (crescents) by performing object detection using learning, corrects the object box through a post-processing algorithm, and uses the results to obtain visual acuity information (e.g., For example, visual acuity information (eg, eye refraction) may be calculated by substituting a function (formula) for predicting eye refraction. In some examples, the deep learning network may be a Single Shot MultiBox Detector (SSD), YOLOv3 or EfficientDet, but may also include other kinds of object detection networks.

The Internet network 130 may connect the user device 110 and the server 120 to communicate with each other through wired or wireless communication. In some examples, Internet network 130 is a Public Switched Telephone Network (PSTN), x Digital Subscriber Line (xDSL), Rate Adaptive DSL (RADSL), Multi Rate DSL (MDSL), Very High Speed DSL (VDSL). ), UADSL (Universal Asymmetric DSL), HDSL (High Bit Rate DSL), and may include various wired communication systems such as local area network (LAN). Also, in some examples, the Internet network 130 is a Code Division Multi Access (CDMA), Time Division Multi Access (TDMA), Frequency Division Multi Access (FDMA), Orthogonal Frequency Division Multi Access (OFDMA), Single Carrier-SCFDMA (SCFDMA) FDMA) and other systems. In addition, in some examples, the Internet network 130 may be configured regardless of its communication aspects, such as wired and wireless, and various communication networks such as a personal area network (PAN) and a wide area network (WAN). may include Also, in some examples, the Internet network 130 may be a well-known World Wide Web (WWW), and a wireless transmission used for short-range communication, such as Infrared Data Assoication (IrDA) or Bluetooth. may include technology.

FIG. 2 is a block diagram illustrating the configuration of the user device 110 and/or the server 120 of the vision test apparatus 100 according to an embodiment of the present invention.

The user device 110 and/or the server 120 may include an input unit 1210 , a control unit 1220 , an output unit 1230 , and a transceiver 1240 in common. The user device 110 may further include a lighting unit 111 and a photographing unit 112 for photographing the eyeball.

The input unit 1210 may include a keyboard, a keypad, and the like, and through this, input of various commands or input of setting values is possible. The output unit 1230 may include a monitor, a display, and the like, and through this, various screen configurations or output of result values are possible. In some examples, the input unit 1210 and the output unit 1230 may include one touch screen.

The control unit 1220 may include a central processing unit 1221 and a memory 1222 , and a program code 1223 for a main operation of the present invention may be stored in the memory 1222 . That is, the control unit 1220 executes the program code 1223 stored in the memory 1222 through the central processing unit 1221 , and accordingly controls the operation of the user device 110 and/or the server 120 . have.

In some examples, the memory 1222 may be a flash memory type, a hard disk type, a multimedia card micro type, or a card type memory (eg SD or XD memory). etc.), Random Access Memory (RAM), Static Random Access Memory (SRAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read-Only Memory (PROM), It may include at least one type of storage medium among a magnetic memory, a magnetic disk, and an optical disk.

The transceiver 1240 is used for receiving and transmitting wired/wireless signals, and transmits a signal received from the Internet network 130 to the controller 1220 , and outputs a signal generated by the controller 1220 to the Internet network 130 . can do. In some examples, transceiver 1240 may include a wired/wireless Internet module for network connection. As wireless Internet technologies, wireless LAN (WLAN) (Wi-Fi), wireless broadband (Wibro), World Interoperability for Microwave Access (Wimax), High Speed Downlink Packet Access (HSDPA), etc. may be used. As the wired Internet technology, Digital Subscriber Line (XDSL), Fibers to the home (FTTH), Power Line Communication (PLC), or the like may be used. The transceiver 1240 according to the embodiments of the present disclosure may additionally include a short-range communication module to transmit/receive data to/from the user device 110 and other user devices 110 including the short-distance communication module located in a relatively short distance from the user device 110 . Bluetooth, Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra Wideband (UWB), ZigBee, etc. may be used as short range communication technologies.

Although the user device 110 and the server 120 are described together here, as described above, the user device 110 may be a smartphone, a tablet, a notebook computer, or a desktop, and the server 120 is used for processing large data in general. It may be a computer. In some examples, the subject method of the present invention is primarily performed on the server 120 , but may also be performed on the user device 110 .

3 is a block diagram showing the configuration of the program code 1223 stored in the memory 1222 of the user device 110 and/or the server 120 of the vision test apparatus 100 according to an embodiment of the present invention. .

As shown in FIG. 3 , the program code 1223 stored in the memory 1222 of the user device 110 and/or the server 120 among the vision test apparatus 100 according to the embodiment of the present invention is a super -Resolution Generative Adversarial Network) module 1224 , an object detection module 1225 , and a vision information calculation function module 1226 may be included.

The SRGAN module 1224 may obtain a super-resolved (SR) image by receiving the pupil image received from the user device 110 . In some examples, the SRGAN module 1224 may output an SR image by removing noise from the received pupil image and increasing the resolution using the SRGAN algorithm that has been trained.

In some examples, the SRGAN module 1224 includes a generator network that receives a noise image and generates a fake pupil image, and a discriminator network that receives a real pupil image and a fake pupil image and makes a prediction, but includes a generator network and a disk The reminator network performs learning alternately with each other, so that the performance of the generator network and the discreminator network can be further improved.

In some examples, the SRGAN module 1224 uses perceptual similarity instead of adversarial loss and pixel space similarity to train a generator network to output the SR imaged content loss. By providing a perceptual loss function including ), it is possible to obtain a clearer SR image.

For example, when a low-resolution pupil image is input to the generator network, the generator network performs a prediction, and a loss occurs between the prediction and the high-resolution pupil image. By inputting this loss back to the generator network, the generator network may be updated. At this time, in calculating the loss, the performance of the generator network is improved by providing the above-described perceptual loss function to describe the characteristic part of the image, rather than learning the similarity in units of pixels like the conventional Mean Squared Error (MSE). to improve

The object detection module 1225 may receive the image SR by the SRGAN module 1224 and obtain result box data. That is, the object detection module 1225 may perform object detection using deep learning to analyze sizes and positions of the iris, pupil, and pupil reflection patterns, and may correct the object box through a post-processing algorithm. As described above, the object detection module 1225 may use at least one of a Single Shot MultiBox Detector (SSD), YOLOv3, and EfficientDet, but the present invention is not limited thereto. is also possible

The eyesight information calculation function module 1226 may receive result box data from the object detection module 1225 and calculate eyesight information. In some examples, the visual acuity information calculation function module 1226 may include a function (formula) for predicting the degree of refraction of the eyeball, thereby calculating the diopter (degree of refraction) of the eyeball.

4 is a flowchart illustrating an eyesight examination method according to an embodiment of the present invention.

As shown in FIG. 4 , the vision test method according to an embodiment of the present invention includes a first photographing step ( S1 ) through the user device 110 , a first image analysis step ( S2 ) through the server 120 , and , may include a second photographing step (S3) through the user device 110, a second image analysis step (S4) through the server 120, and a result derivation step (S5) through the user device 110 have. Here, the first image analysis step (S2) and the second image analysis step (S4) may be performed in the server 120, and the server 120 includes the above-described SRGAN module 1224, object detection module 1225 and Image analysis is performed through the program code 1223 including the vision information calculation function module 1226 . In addition, the result of image analysis is transmitted from the server 120 to the user device 110, and the user device 110 displays it (result derivation step S5). In some examples, the first image analysis may be an analysis of a pupil image on a 90 degree axis (a pupil image taken by setting the user device 110 at 90 degrees with respect to the ground), and the second image analysis may be an analysis of the pupil image on the user device 110 . ) may be an analysis of a pupil image on a 180-degree axis (a pupil image photographed by setting the user device 110 at 180 degrees with respect to the ground).

5 is a flowchart illustrating an image analysis method among the visual acuity testing methods according to an embodiment of the present invention.

As shown in FIG. 5 , the image analysis method among the vision test methods according to an embodiment of the present invention includes an image receiving step (S21), an SRGAN operation step (S22), an SR image acquisition step (S23), an object detection step ( S24), the result box data acquisition step (S25), the operation step of the eyesight information calculation function (S26), and the eyesight information acquisition step (S27) may be included.

In the image receiving step S21 , for example, a pupil image photographed from the user device 110 is received through the Internet network 130 . In some examples, the image receiving step may be performed by receiving a pupil image from the user device 110 including the lighting unit 111 and the photographing unit 112 , and capable of wired or wireless communication.

In the SRGAN operation step (S22) and the SR image acquisition step (S23), the SRGAN module 1224 removes noise from the pupil image using the SRGAN algorithm that has been trained and increases the resolution to output the SR image. In some examples, the SRGAN module 1224 includes a generator network that receives a noise image and generates a fake pupil image, and a discriminator network that receives a real pupil image and a fake pupil image and makes a prediction, but includes a generator network and a disk The reminator network can perform learning by taking turns with each other. In addition, the SRGAN module 1224 may provide a perceptual loss function including a content loss using perceptual similarity instead of adversarial loss and pixel space similarity to train a generator network that outputs an SR image.

In the object detection step (S24) and the result box data acquisition step (S25), the object detection module 1225 performs object detection using deep learning to analyze the size and position of the iris, pupil, and pupil reflection patterns, and post-processing You can calibrate the object box through an algorithm. In some examples, the object detection module 1225 may use at least one of a Single Shot MultiBox Detector (SSD), YOLOv3, or EfficientDet.

In the vision information calculation function operation step S26 and the vision information acquisition step S27 , the vision information calculation function module 1226 receives and processes the result box data to calculate vision information. In some examples, the vision information may include eye refraction as described above.

6A and 6B are block diagrams illustrating a learning algorithm using a Generative Adversarial Network (GAN) and a Super-Resolution Using a Generative Adversarial Network (SRGAN) in an image analysis method according to an embodiment of the present invention.

As shown in FIG. 6A , the GAN module includes a generator network that generates a fake image by receiving a noise image, and a discriminator network that receives real images and fake images and makes predictions, a generator network and a discriminator network By cross-progressing mutual learning, the performance of the generator network and the discriminator network is improved over the learning time. This GAN module _{learns to minimize and maximize the parameter θ D} of the discriminator network and the parameter θ _G of the generator network, respectively, as shown below.

Here, I ^LR is a low-resolution input image, and I ^HR is a high-resolution output image.

In addition, as shown in FIG. 6B , the SRGAN module 1224 also includes a generator network and a discriminator network, and learns like the general GAN module described above. That is, the discriminator network learns to distinguish the original high-resolution image from the image created by SR, so that the generator generates an image that is difficult to distinguish from the high-resolution image.

Meanwhile, to learn the generator network, B residual blocks are used in the same shape. Here, B=16 may be used, and 16 blocks are used. The residual block contains two convolutional layers with a 3 x 3 kernel, and batch-normalization is applied and ParametricReLU is used as the activation function. . The delimiter uses the LeakyReLU activation function. Eight convolutional layers using a 3 x 3 kernel are used like the structure of the VGG network, and the number of feature maps increases by a factor of 2 from 64 to 512. 512 feature maps enter the density layer and go through sigmoid activation to determine whether the input is an actual image or an SR image.

For example, the SRGAN module 1224 performs SRGAN learning based on a database in which a high-resolution original image and a reduced low-resolution image matching it are stored together during training, so that a clearer image with higher resolution can be obtained. Although SRGAN learning has been described in the embodiment of the present invention, it is not limited to SRGAN learning. Specifically, the type of deep learning learning-based SR (Super-Resolution) is not limited to SRGAN, and may be SRCNN (Super-Resolution Convolutional Neural Network). That is, in the present invention, it is also possible to perform learning with a convolutional neural network (CNN)-based SR learning algorithm. Accordingly, since the embodiment of the present invention performs SRGAN learning or SRCNN learning, a clear pupil image with higher resolution can be obtained.

7 is a block diagram illustrating a learning algorithm using a Single Shot MultiBox Detector (SSD) and YOLO in an image analysis method according to an embodiment of the present invention.

SSD takes advantage of the fact that the feature map gets smaller and smaller as it undergoes convolution operation. With a default box, small objects are detected in large feature maps and large objects are detected in small feature maps. can be detected. Also, unlike YOLO, the category and offset of an object can be predicted only by convolution operation without using a fully connected layer. Therefore, the amount of computation is smaller than that of YOLO, showing better speed.

As shown in Figure 7, the SSD model adds multiple feature layers at the end of the base network, which predicts the offsets and aspect ratios of default boxes of different scales and their associated confidence. For example, a 300 x 300 input size SSD outperforms its 448 x 448 YOLO counterpart in terms of accuracy and is also fast.

Therefore, in an embodiment of the present invention, the object detection module 1225 performs object detection using deep learning to accurately analyze the size and position of the iris, pupil, and pupil reflection patterns, and accurately analyzes the object box through a post-processing algorithm. By correcting it, it is possible to more accurately calculate the visual acuity information.

What has been described above is only one embodiment for implementing the vision test apparatus, method, and program stored in the computer-readable storage medium according to the present invention, and the present invention is not limited to the above-described embodiment, and the following claims are made. As claimed in the scope, without departing from the gist of the present invention, it will be said that the technical spirit of the present invention exists to the extent that various modifications can be made by anyone with ordinary knowledge in the field to which the invention pertains.

Claims (18)

receiving a pupil image; inputting the pupil image into an SRGAN (Super-Resolution Generative Adversarial Network) module; obtaining an SR (Super-Resolved) image from the SRGAN module; inputting the SR image to the object detection module; obtaining result box data from the object detection module; inputting the result box data into an eyesight information calculation function module; and The vision test method, comprising the step of acquiring and transmitting the vision information from the vision information calculation function module. The method of claim 1, The pupil image receiving step includes an illumination unit and a photographing unit, and is made by receiving a pupil image from a user device capable of wired or wireless communication. The method of claim 1, The SRGAN module uses the learned SRGAN algorithm to remove noise from the received pupil image and to increase the resolution to output the SR image. The method of claim 1, The SRGAN module includes a generator network that receives a noise image and generates a fake pupil image, and a discriminator network that receives and predicts an actual pupil image and a fake pupil image, but the generator network and the discriminator network are alternated with each other. How to do learning, eye exam. 5. The method of claim 4, The SRGAN module uses perceptual similarity instead of adversarial loss and pixel space similarity to train a generator network that outputs the SR image. A method of examining vision, providing a perceptual loss function. The method of claim 1, The object detection module performs object detection using deep learning to analyze the size and position of the iris, pupil, and pupil reflection patterns, and corrects the object box through a post-processing algorithm. The method of claim 1, The object detection module uses at least one of a Single Shot MultiBox Detector (SSD), YOLOv3, or EfficientDet. The method of claim 1, The step of transmitting the vision information is performed by transmitting the vision information to a user device capable of wired or wireless communication. The method of claim 1, The visual acuity information includes an eye refractive index. receiving a pupil image; inputting the pupil image to the SRGAN (Super-Resolution Generative Adversarial Network) module; obtaining an SR (Super-Resolved) image from the SRGAN module; inputting the SR image to the object detection module; obtaining result box data from the object detection module; inputting the result box data into the vision information calculation function module; and A computer program stored in a computer-readable storage medium for eye examination, comprising the operation of acquiring and transmitting vision information from a vision information calculation function module. a user device that captures and transmits a pupil image; and An SRGAN (Super-Resolution Generative Adversarial Network) module that receives the pupil image received from the user device and acquires an SR (Super-Resolved) image, and an object detection module that receives the SR image and acquires result box data; and a server having an eyesight information calculation function module that receives result box data and calculates vision information, wherein the server transmits the calculated vision information to the user device. 12. The method of claim 11, The user device includes a lighting unit and a photographing unit, and transmits the photographed pupil image to the server through a wired or wireless communication method. 12. The method of claim 11, The SRGAN module removes noise from the received pupil image using the learned SRGAN algorithm and outputs the SR image by increasing the resolution. 12. The method of claim 11, The SRGAN module includes a generator network that receives a noise image and generates a fake pupil image, and a discriminator network that receives and predicts an actual pupil image and a fake pupil image, but the generator network and the discriminator network are alternated with each other. A vision test device that performs learning. 15. The method of claim 14, The SRGAN module uses perceptual similarity instead of adversarial loss and pixel space similarity to train a generator network that outputs the SR image. An eye exam device that provides a perceptual loss function. 12. The method of claim 11, The object detection module performs object detection using deep learning to analyze the size and position of the iris, pupil, and pupil reflection patterns, and corrects the object box through a post-processing algorithm. 12. The method of claim 11, The object detection module uses at least one of a Single Shot MultiBox Detector (SSD), YOLOv3, or EfficientDet. 12. The method of claim 11, The visual acuity information includes an eye refractive index.

Images