Apparatus and method for image restoration based on machine learning

Abstract

An image restoration method according to one disclosed embodiment is a method performed in an image restoration device comprising a processor and a memory storing one or more programs executed by the processor, comprising the steps of: calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image; applying the predicted color vector to the scanned image to produce a color-corrected image; calculating predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image; and removing the predicted noise from the color-corrected image to produce a final correction image.

Description

Technology Field

[0001] An embodiment of the present invention relates to machine learning-based image restoration technology. Background Technology

[0003] As storing information as digital documents becomes more common, the technology of scanning printed information and storing it digitally is being increasingly utilized. In order to accurately store the information contained in a scanned image, it is important to restore the scanned image to be as close as possible to the original image.

[0004] Conventional techniques for restoring scanned images to their original state have focused solely on removing single types of degradation, such as Gaussian noise and blurriness. Furthermore, these techniques correct for degradation—such as physical wrinkling, shadows, and color shifts resulting from program processing—using only traditional image processing methods; consequently, they have suffered from poor processing efficiency regarding degradation types that are not predefined.

[0005] Accordingly, there is a need for scan image restoration technology that can be applied to new types of image degradation other than Gaussian noise and pre-set image degradation types. Prior art literature

[0007] Korean Patent Publication No. 10-2123611 (June 29, 2020) The problem to be solved

[0008] The objective of the present invention is to provide a scan image restoration technology applicable to novel image degradation types other than Gaussian noise and preset image degradation types. means of solving the problem

[0010] An image restoration method according to one disclosed embodiment is a method performed in an image restoration device comprising a processor and a memory storing one or more programs executed by the processor, comprising the steps of: calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image; applying the predicted color vector to the scanned image to produce a color-corrected image; calculating predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image; and removing the predicted noise from the color-corrected image to produce a final correction image.

[0011] The step of calculating the predicted color vector may include the step of calculating the predicted color vector by predicting the color distribution of the original image from the scanned image, the step of calculating the original color vector representing the color distribution from the original image, and the step of training the neural network of the first learning model that calculates the predicted color vector based on the predicted color vector and the original color vector.

[0012] The step of training the neural network of the first learning model may include the step of calculating a first loss function value representing the difference between the predicted color vector and the original color vector, and the step of adjusting the factors (parameters) constituting the neural network of the first learning model so as to reduce the first loss function value.

[0013] The first loss function above may be a function in which the value of the first loss function is calculated to be smaller as the difference between the predicted color vector and the original color vector becomes smaller.

[0014] The above first loss function may be equal to Equation 1 below.

[0015] [Mathematical Formula 1]

[0016]

[0017] Loss1: First loss function value

[0018] : Original color vector

[0019] : Predicted color vector

[0020] The step of calculating the predicted noise may include: a step of calculating the predicted noise based on the color-corrected image and preset noise; a step of calculating the predicted noise by inputting the color-corrected image into the second learning model configured to add incremental noise; a step of comparing the original image and the color-corrected image to calculate the difference noise as theoretical noise; and a step of training the neural network of the second learning model that calculates the predicted noise based on the predicted noise and the theoretical noise.

[0021] The step of training the neural network of the second learning model may include the step of calculating a second loss function value representing the difference between the prediction noise and the theoretical noise, and the step of adjusting the factors (parameters) constituting the neural network of the second learning model so as to reduce the second loss function value.

[0022] The second loss function above may be a function in which the value of the second loss function is calculated to be smaller as the difference between the prediction noise and the theoretical noise becomes smaller.

[0023] The above second loss function may be as shown in Equation 2 below.

[0024] [Mathematical Formula 2]

[0025]

[0026] Loss2: Second loss function value

[0027] : Theoretical noise

[0028] : Image with gradual noise added

[0029] : Color-corrected image

[0030] An image restoration device according to one disclosed embodiment is an image restoration device comprising a processor and a memory for storing one or more programs executed by the processor, wherein the one or more programs include a command for calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image, a command for calculating a color-corrected image by applying the predicted color vector to the scanned image, a command for calculating predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image, and a command for producing a final corrected image by removing the predicted noise from the color-corrected image.

[0031] A computer program according to one disclosed embodiment is a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program comprises one or more instructions, and when the instructions are executed by a computing device having one or more processors, the computing device performs the steps of: calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image; applying the predicted color vector to the scanned image to produce a color-corrected image; calculating predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image; and removing the predicted noise from the color-corrected image to produce a final correction image. Effects of the invention

[0033] According to one disclosed embodiment, in restoring a scanned image to a high-quality image, it is possible to remove a novel type of image degradation other than Gaussian noise and a preset type of image degradation.

[0034] The effects of the present invention are not limited to those mentioned above, and other effects of the present specification not mentioned may be understood from the following description and will be more clearly understood by the embodiments of the present specification. Furthermore, it will be readily apparent that the effects of the present specification can be realized by the means and combinations thereof set forth in the claims. Brief explanation of the drawing

[0036] The present disclosure can be easily understood from the combination of the following detailed description and the accompanying drawings, where reference numerals denote structural elements. FIG. 1 is a block diagram showing the configuration of a computing device that performs restoration of a scanned image according to one embodiment. FIG. 2 is a drawing showing an example of degradation that may occur on a scanned image according to one embodiment. FIG. 3 is a schematic diagram illustrating a method for restoring a scanned image according to one embodiment. FIG. 4 is a flowchart illustrating a method for restoring a scanned image according to one embodiment. FIG. 5 is a schematic diagram illustrating how a computing device removes color degradation of a scanned image according to one embodiment. FIG. 6 is a schematic diagram illustrating how a computing device removes noise (degradation other than color) from a scanned image according to one embodiment. Specific details for implementing the invention

[0037] Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The following detailed description is provided to facilitate a comprehensive understanding of the methods, apparatuses, and / or systems described herein. However, this is merely illustrative and the present invention is not limited thereto.

[0038] In describing the embodiments of the present invention, detailed descriptions of known technologies related to the present invention are omitted if it is determined that such detailed descriptions may unnecessarily obscure the essence of the present invention. Furthermore, the terms described below are defined in consideration of their functions within the present invention, and these may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification. Terms used in the detailed description are intended merely to describe the embodiments of the present invention and should not be limiting in any way. Unless explicitly stated otherwise, expressions in the singular form include the meaning of the plural form. In this description, expressions such as "include" or "comprise" are intended to refer to certain characteristics, numbers, steps, actions, elements, parts thereof, or combinations thereof, and should not be interpreted to exclude the existence or possibility of one or more other characteristics, numbers, steps, actions, elements, parts thereof, or combinations thereof other than those described.

[0039] Additionally, terms including ordinal numbers, such as 'first' or 'second', may be used to describe various components, but said components should not be limited by said terms. Such terms may be used for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component.

[0040] Furthermore, when it is said that one configuration is "connected" to another configuration, this includes not only cases where they are "directly connected," but also cases where they are "connected with another configuration in between."

[0042] FIG. 1 is a block diagram showing the configuration of a computing device that performs the restoration of a scanned image according to one embodiment. In the illustrated embodiment, each component may have different functions and capabilities in addition to those described below, and may include additional components in addition to those described below.

[0043] The illustrated computing environment (10) includes a computing device (12). In one embodiment, the computing device (12) can perform the scan image restoration method (S100) described in FIGS. 2 to 6.

[0044] The computing device (12) includes at least one processor (14), a computer-readable storage medium (16), and a communication bus (18). The processor (14) may enable the computing device (12) to operate according to exemplary embodiments described below. For example, the processor (14) may execute one or more programs stored in the computer-readable storage medium (16). The one or more programs may include one or more computer-executable instructions, and the computer-executable instructions may be configured to enable the computing device (12) to perform operations according to exemplary embodiments when executed by the processor (14).

[0045] A computer-readable storage medium (16) is configured to store computer-executable instructions or program code, program data and / or other suitable forms of information. A program (20) stored in the computer-readable storage medium (16) includes a set of instructions executable by a processor (14). In one embodiment, the computer-readable storage medium (16) may be memory (volatile memory such as random access memory, non-volatile memory, or a suitable combination thereof), one or more magnetic disk storage devices, optical disk storage devices, flash memory devices, or other forms of storage media that are accessed by a computing device (12) and capable of storing desired information, or a suitable combination thereof.

[0046] The communication bus (18) interconnects various other components of the computing device (12), including the processor (14) and the computer-readable storage medium (16).

[0047] The computing device (12) may also include one or more input / output interfaces (22) and one or more network communication interfaces (26) that provide interfaces for one or more input / output devices (24). The input / output interfaces (22) and network communication interfaces (26) are connected to a communication bus (18). The input / output devices (24) may be connected to other components of the computing device (12) through the input / output interfaces (22). An exemplary input / output device (24) may include an input device such as a pointing device (such as a mouse or trackpad), a keyboard, a touch input device (such as a touchpad or touchscreen), a voice or sound input device, various types of sensor devices and / or imaging devices, and / or an output device such as a display device, a printer, a speaker and / or a network card. An exemplary input / output device (24) may be included inside the computing device (12) as a component constituting the computing device (12), or it may be connected to the computing device (12) as a separate device distinct from the computing device (12).

[0049] FIG. 2 is a diagram illustrating examples of degradation that may occur on a scanned image according to one embodiment. Referring to FIG. 2, a scanned image obtained by scanning an image appearing on a ground surface may have six types of degradation. However, this is merely an example, and the degradation that may appear in a scanned image is not limited thereto, nor is the degradation that can be removed by the scanned image restoration method of the present invention limited to the above examples.

[0050] The first scanned image (10S) may include deterioration of external noise (11S), internal noise (12S), and a bleed-through effect (13S).

[0051] External noise (11S) may occur due to the introduction of foreign substances during processes such as printing, scanning, and preservation. External noise (11S) may appear as dots or localized stains in contrast to an image (11O) without external noise. Internal noise (12S) may occur during the process of irradiating light (laser) during scanning. Internal noise (12S) may appear as a curved shape or a linear laser pattern in contrast to an image (12O) without internal noise. Back-side reflection effect (13S) may occur due to the phenomenon where light for scanning is transmitted and reflected by the image on the back. The back-side reflection effect (13S) may appear as a phenomenon where the image on the back of the page being scanned appears blurry when scanned together with the image (13O) without back-side reflection.

[0052] The second scanned image (10S) may include texture distortion (21S), halftone pattern (22S), and color transition (23S) degradation.

[0053] Texture distortion (21S) occurs due to causes similar to external noise (11S) but may appear throughout the scanned image. Texture distortion (21S) may appear as continuous small dots and lines compared to an image without texture distortion (21O). Halftone patterns (22S) may be caused by distortion during the pigment distribution process when the original image is printed and then scanned. Compared to an image without halftones (22O), the difference between the two colors may appear weaker. Color change (23S) may be a change in color, saturation, etc., throughout during scanning or preservation compared to an image without color change (23O).

[0055] FIG. 3 is a schematic diagram illustrating a method for restoring a scanned image according to one embodiment, and FIG. 4 is a flowchart illustrating a method for restoring a scanned image according to one embodiment. The method illustrated in FIG. 4 may be performed, for example, by the computing device (12) described above. Although the method is described in the illustrated flowchart in a plurality of steps, at least some of the steps may be performed in a different order, combined with other steps and performed together, omitted, divided into detailed steps, or performed with one or more steps not illustrated added.

[0056] Referring to FIGS. 3 and 4, a method for restoring a scanned image (S100) may include the step of generating a color-corrected image with color degradation removed by applying a predicted color vector to the scanned image (S110), and the step of generating a final corrected image by removing predicted noise (non-color degradation) from the color-corrected image (S120).

[0057] A computing device (12) can perform a method (S100) for restoring a scanned image through a machine learning model of an artificial neural network structure. The artificial neural network structure of the machine learning model may include one or more layers and neurons that perform pre-set functions. The computing device (12) can train a machine learning model for restoring a scanned image by adjusting factors (parameters) that constitute the layers and neurons.

[0058] In step S110, the computing device (12) can generate a color-corrected image with color degradation removed by applying a predicted color vector to the scanned image. Specific details are as follows.

[0059] The computing device (12) can acquire a scanned image. The computing device (12) may acquire a scanned image by scanning a paper that displays an image directly, including a scanner, or the computing device (12) may acquire a scanned image in the form of data from an external source. Additionally, the computing device (12) can acquire data of the original image that outputs the scanned image to the paper.

[0060] The computing device (12) can calculate a color vector (hereinafter, predicted color vector) that is predicted to be distributed in the original image from the scanned image. The computing device (12) can generate a color-corrected image with color degradation removed by adjusting the color distribution of the scanned image to the predicted color vector. More specific details regarding the computing device (12) calculating the predicted color vector and obtaining the color-corrected image will be described later through FIG. 5.

[0061] In step S120, the computing device (12) can generate a final corrected image by removing predicted noise (non-color degradation) from the color-corrected image. Specific details are as follows.

[0062] The computing device (12) can calculate noise (hereinafter, predicted noise) that is predicted to have been added from the original image to the color-corrected image from which color degradation has been removed. The computing device (12) can generate a corrected image (hereinafter, final corrected image) by removing the predicted noise from the color-corrected image. More specific details regarding the computing device (12) calculating the predicted noise and obtaining the final corrected image will be described later through FIG. 6.

[0064] FIG. 5 is a schematic diagram illustrating how a computing device removes color degradation of a scanned image according to one embodiment.

[0065] Referring to FIG. 5, a method for generating a color-corrected image (S110) may include the step of calculating a predicted color vector from a scanned image (S112-1), the step of extracting a color vector (hereinafter, original color vector) representing the color distribution of the original image from the original image (S112-2), the step of adjusting a first learning model that calculates the predicted color vector so as to minimize the difference between the predicted color vector and the original color vector (S114), and the step of applying the predicted color vector calculated according to the adjusted first learning model to the scanned image (S116).

[0066] In step S112-1, the computing device (12) can produce a predicted color vector from the scanned image. The specific details are as follows. The computing device (12) can produce a predicted color vector by passing the acquired scanned image through a color encoder layer. The predicted color vector may represent the color distribution of the original image corresponding to the scanned image. The color encoder layer that produces the predicted color vector may include a plurality of neurons as part of an artificial neural network, and its operation may be controlled by a parameter constituting the plurality of neurons.

[0067] In step S112-2, the computing device (12) can extract a color vector (hereinafter referred to as the original color vector) representing the color distribution of the original image from the original image. The specific details are as follows. The computing device (12) can acquire an original image corresponding to the input scanned image. The computing device (12) can classify the acquired original image into one or more colors and extract an original color vector having the average value and the variance value of the distribution of each color as elements.

[0068] In step S114, the computing device (12) can adjust a first learning model that calculates a predicted color vector such that the difference between the predicted color vector and the original color vector is minimized. The specific details are as follows. The computing device (12) can calculate a first loss value representing the difference between the predicted color vector and the original color vector. In an exemplary embodiment, the first loss function (Loss1) that calculates the first loss value may be as shown in [Equation 1] below.

[0069] [Mathematical Formula 1]

[0070]

[0071] Loss1: First loss function value

[0072] : Original color vector

[0073] : Predicted color vector

[0074] The computing device (12) can adjust the factors of the color encoder layer that generate the predicted color vector so as to minimize the first loss value. As the factors of the color encoder layer are adjusted, the computing device (12) can be trained to produce a predicted color vector similar to the original color vector.

[0075] In step S116, the computing device (12) can apply the predicted color vector calculated according to the adjusted first learning model to the scanned image. The specific details are as follows. The computing device (12) can obtain the predicted color vector of the scanned image by using the scanned image as input to a learned color encoder layer. The computing device (12) can obtain a color-corrected image with color degradation removed by changing the color distribution of the scanned image based on the predicted color vector.

[0077] FIG. 6 is a schematic diagram illustrating how a computing device removes noise (degradation other than color) from a scanned image according to one embodiment.

[0078] Referring to FIG. 6, a method for generating a final corrected image (S120) may include a step of calculating prediction noise from a color corrected image (S122), a step of adjusting a second learning model that calculates prediction noise so that the difference between the prediction noise and the theoretical noise is minimized (S124), and a step of removing the prediction noise calculated according to the adjusted second learning model from the color corrected image (S126).

[0079] In step S122, the computing device (12) can calculate predicted noise from the color-corrected image. The specific details are as follows. The computing device (12) can calculate predicted noise by passing the color-corrected image through a noise predictor layer. In an exemplary embodiment, the computing device (12) can obtain predicted noise by using Gaussian noise based on the color-corrected image and the original image as input to the noise predictor layer.

[0080] In an exemplary embodiment, the noise prediction layer may implement a diffusion model. The diffusion model may be a generative model that learns to restore the original data by gradually adding noise to the data and then removing it. The computing device (12) may gradually strengthen or weaken the intensity of the prediction noise through the diffusion model and may calculate the prediction noise by applying multiple noises in a superposition manner.

[0081] In step S124, the computing device (12) can adjust a second learning model that calculates the predicted noise so that the difference between the predicted noise and the theoretical noise is minimized. The specific details are as follows. The computing device (12) can obtain a Gaussian distribution value for the original image as the theoretical noise. The theoretical noise may represent the noise contained in the color-corrected image compared to the original image. For example, the theoretical noise may be the noise that is different when comparing the original image and the color-corrected image.

[0082] The computing device (12) can calculate a second loss value representing the difference between the predicted noise and the theoretical noise. In an exemplary embodiment, the second loss function (Loss2) for calculating the second loss value may be as shown in the following Equation 2.

[0083] [Mathematical Formula 2]

[0084]

[0085] Loss2: Second loss function value

[0086] : Theoretical noise

[0087] : Image with progressively added noise

[0088] : Color-corrected image

[0089] : Predicted color vector

[0090] : Factor for distinguishing multiple noise generation

[0091] In this case, the gradual noise can be Gaussian noise.

[0092] The computing device (12) can adjust the factors of the noise prediction layer that generate prediction noise so that the second loss value is minimized. As the factors of the noise prediction layer are adjusted, the computing device (12) can be trained to produce prediction noise similar to theoretical noise. That is, as the factors of the noise prediction layer are adjusted, the computing device can be trained so that the image from which noise has been removed from the production correction image (final correction image) is similar to the original image.

[0093] In step S126, the computing device (12) can remove the predicted noise calculated according to the adjusted second learning model from the color-corrected image. The specific details are as follows. The computing device (12) can obtain the predicted noise of the color-corrected image by inputting the color-corrected image to a learned noise prediction layer. The computing device (12) can produce a final restored image by removing the predicted noise expected to be present in the color-corrected image from the color-corrected image.

[0095] According to one disclosed embodiment, in restoring a scanned image to a high-quality image, it is possible to remove a novel type of image degradation other than Gaussian noise and a preset type of image degradation.

[0096] The effects of the present invention are not limited to those mentioned above, and other effects of the present specification not mentioned may be understood from the following description and will be more clearly understood by the embodiments of the present specification. Furthermore, it will be readily apparent that the effects of the present specification can be realized by the means and combinations thereof set forth in the claims.

[0098] Although representative embodiments of the present invention have been described in detail above, those skilled in the art will understand that various modifications can be made to the above-described embodiments without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be defined by the claims set forth below as well as equivalents thereof. Explanation of the symbols

[0100] 10: Computing Environment 12: Computing device 14: Processor 16: Computer-readable storage media 18: Communication bus 20: Program 22: Input / Output Interface 24: Input / Output Devices 28: Network communication interface

Claims

Claim 1 delete Claim 2 A method for restoring an image, comprising: a processor; and a memory for storing one or more programs executed by the processor, wherein the method comprises: a step of calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image; a step of applying the predicted color vector to the scanned image to produce a color-corrected image; a step of calculating predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image; and a step of removing the predicted noise from the color-corrected image to produce a final correction image, wherein the step of calculating the predicted color vector comprises: a step of calculating a predicted color vector by predicting the color distribution of the original image from the scanned image; a step of calculating an original color vector representing the color distribution from the original image; and a step of training a neural network of the first learning model to calculate the predicted color vector based on the predicted color vector and the original color vector. Claim 3 An image restoration method according to claim 2, wherein the step of training the neural network of the first learning model comprises: a step of calculating a first loss function value representing the difference between the predicted color vector and the original color vector; and a step of adjusting a factor (parameter) constituting the neural network of the first learning model so as to reduce the first loss function value. Claim 4 An image restoration method according to claim 3, wherein the first loss function is a function in which the value of the first loss function is calculated to be smaller as the difference between the predicted color vector and the original color vector becomes smaller. Claim 5 In claim 4, the first loss function is an image restoration method as shown in Equation 1 below. [Equation 1] Loss1: First loss function value : Original color vector : Predicted color vector Claim 6 A method for restoring an image, comprising: a processor; and a memory for storing one or more programs executed by the processor, wherein the method comprises: a step of calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image; a step of applying the predicted color vector to the scanned image to produce a color-corrected image; a step of calculating a predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image; and a step of removing the predicted noise from the color-corrected image to produce a final correction image, wherein the step of calculating the predicted noise comprises: a step of calculating the predicted noise based on the color-corrected image and a preset noise; a step of inputting the color-corrected image into the second learning model set to add progressive noise to produce the predicted noise; a step of comparing the original image and the color-corrected image to calculate the difference noise as theoretical noise; and a step of training a neural network of the second learning model that calculates the predicted noise based on the predicted noise and the theoretical noise. Claim 7 An image restoration method according to claim 6, wherein the step of training the neural network of the second learning model comprises: a step of calculating a second loss function value representing the difference between the prediction noise and the theoretical noise; and a step of adjusting a factor (parameter) constituting the neural network of the second learning model so as to reduce the second loss function value. Claim 8 An image restoration method according to claim 7, wherein the second loss function is a function in which the value of the second loss function is calculated to be smaller as the difference between the predicted noise and the theoretical noise becomes smaller. Claim 9 In claim 8, the second loss function is an image restoration method as shown in Equation 2 below. [Equation 2] Loss2: Second loss function value : Theoretical noise : Image with gradual noise added : Color-corrected image Claim 10 delete Claim 11 An image restoration device comprising: a processor; and a memory for storing one or more programs executed by the processor, wherein the one or more programs include: a command for calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image; a command for calculating a color-corrected image by applying the predicted color vector to the scanned image; a command for calculating a predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image; and a command for producing a final correction image by removing the predicted noise from the color-corrected image, wherein the command for calculating the predicted color vector includes: a command for calculating a predicted color vector by predicting the color distribution of the original image from the scanned image; a command for calculating an original color vector representing the color distribution from the original image; and a command for training a neural network of the first learning model to calculate the predicted color vector based on the predicted color vector and the original color vector. Claim 12 An image restoration device according to claim 11, wherein a command for training the neural network of the first learning model comprises: a command for calculating a first loss function value representing the difference between the predicted color vector and the original color vector; and a command for adjusting a factor (parameter) constituting the neural network of the first learning model so as to reduce the first loss function value. Claim 13 An image restoration device according to claim 12, wherein the first loss function is a function in which the value of the first loss function is calculated to be smaller as the difference between the predicted color vector and the original color vector becomes smaller. Claim 14 In claim 13, the first loss function is an image restoration device as shown in Equation 1 below. [Equation 1] Loss1: First loss function value : Original color vector : Predicted color vector Claim 15 An image restoration device comprising: a processor; and a memory for storing one or more programs executed by the processor, wherein the one or more programs include: a command for calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image; a command for calculating a color-corrected image by applying the predicted color vector to the scanned image; a command for calculating a predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image; and a command for producing a final correction image by removing the predicted noise from the color-corrected image, wherein the command for calculating the predicted noise includes: a command for calculating the predicted noise based on the color-corrected image and a preset noise; a command for calculating the predicted noise by inputting the color-corrected image into the second learning model set to add progressive noise; a command for calculating the difference noise as theoretical noise by comparing the original image and the color-corrected image; and a command for training a neural network of the second learning model that calculates the predicted noise based on the predicted noise and the theoretical noise. Claim 16 An image restoration device according to claim 15, wherein a command for training the neural network of the second learning model comprises: a command for calculating a second loss function value representing the difference between the prediction noise and the theoretical noise; and a command for adjusting a factor (parameter) constituting the neural network of the second learning model so as to reduce the second loss function value. Claim 17 An image restoration device according to claim 16, wherein the second loss function is a function in which the value of the second loss function is calculated to be smaller as the difference between the predicted noise and the theoretical noise becomes smaller. Claim 18 In claim 17, the image restoration device, wherein the second loss function is as shown in Equation 2 below. [Equation 2] Loss2: Second loss function value : Theoretical noise : Image with gradual noise added : Color-corrected image Claim 19 A computer program stored on a non-transitory computer-readable storage medium, wherein the computer program comprises one or more instructions, and when the instructions are executed by a computing device having one or more processors, the computing device performs the steps of: calculating a predicted color vector by a first learning model learned to predict the color distribution of an original image from a scanned image; applying the predicted color vector to the scanned image to produce a color-corrected image; calculating predicted noise by a second learning model learned to predict the noise contained in the color-corrected image in contrast to the original image; and removing the predicted noise from the color-corrected image to produce a final correction image, wherein the step of calculating the predicted color vector includes: a step of calculating a predicted color vector by predicting the color distribution of the original image from the scanned image; a step of calculating an original color vector representing the color distribution from the original image; and a step of training a neural network of the first learning model to calculate the predicted color vector based on the predicted color vector and the original color vector.

Images