Method and apparatus for converting images based on artificial intelligence

Abstract

The present invention relates to method and apparatus for converting images based on artificial intelligence, the method comprising: a first step of determining whether data input from a user is an image or a video; a second step of preprocessing an image if the data input from the user is an image; a third step of converting the preprocessed image into a High Dynamic Range (HDR) image using artificial intelligence; a fourth step of reducing the HDR image; and a fifth step of inputting the reduced HDR image to a pre-trained model and converting it into a high-resolution image. [Representative Figure] FIG. 4

Description

METHOD AND APPARATUS FOR CONVERTING IMAGES BASED ON ARTIFICIAL INTELLIGENCE

[0001] The present invention relates to an image conversion system, and more specifically, to method and apparatus for converting images based on artificial intelligence automatically generating and rendering realistic low-light images using images taken in environments with limited light and camera information.

[0002] Images acquired in the evening or in areas with low light are referred to as low-light images, and the dark areas of these low-light images reduce the detail and quality of the images, causing various computer vision algorithms such as object detection or tracking to perform poorly. To solve this problem, various low-light image resolution methods have been proposed in the field of computer vision.

[0003] Especially, there are methods based on histogram equalization using the brightness and contrast of an image, retinex theory, and High Dynamic Range (HDR). In addition, recently, methods such as MSRNet, LLNet, LightenNet, RetinexNet, and MBLLEN using deep learning technology have been proposed. Furthermore, there are also methods such as low-lightGAN and EnlightenGAN that realistically improve low-light images using Generative Adversarial Networks (GAN). Although the images generated by these conventional techniques have the effect of improving low-light images, there are problems in that low-light areas are not completely improved due to unremoved noise, and the colors are forcibly corrected, so the colors are more exaggerated and heterogeneous.

[0004] [Prior Art Documents]

[0005] (Patent Document 1) Korean Patent Registration Publication No. 10-2368677

[0006] (Patent Document 2) Korean Patent Registration Publication No. 10-2252557

[0007] The present specification is devised to solve the above-mentioned problems, and the purpose of the present invention is to provide method and apparatus for converting images based on artificial intelligence capable of correcting a photograph taken in an environment where an extremely small amount of light enters.

[0008] It is another purpose of the present invention to provide method and apparatus for converting images based on artificial intelligence that can be applied and used not only by specific camera vendors but also by any camera vendor.

[0009] According to an embodiment of the present specification to achieve this purpose, an image conversion method based on artificial intelligence according to the present specification includes: a first step of determining whether data input from a user is an image or a video; a second step of preprocessing an image if the data input from the user is an image; a third step of converting the preprocessed image into a High Dynamic Range (HDR) image using artificial intelligence; a fourth step of reducing the HDR image; and a fifth step of inputting the reduced HDR image to a pre-trained model and converting it into a high-resolution image.

[0010] Preferably, the method further includes: a first sub-step of the first step of dividing a video into a plurality of frames if the data input from the user is a video; and a sixth step of performing the second to fifth steps on each frame, and then merging the converted plurality of frames to generate a video.

[0011] Preferably, the method further includes a first sub-step of the fifth step of performing frame interpolation between two consecutive frames.

[0012] Preferably, the preprocessing includes at least one of linearization, normalization, demosaicing, denoising, white balancing, and gamma correction.

[0013] Preferably, the fourth step includes emphasizing contrast of the HDR image and trimming size and rotation of the HDR image.

[0014] Preferably, the third step includes converting the preprocessed image into the HDR image through inverse tone mapping.

[0015] Preferably, the fifth step includes training the model using the reduced HDR image and an original image as a dataset.

[0016] According to another embodiment of the present specification, an image conversion apparatus according to the present specification includes: a memory storing one or more instructions; and a processor executing the one or more instructions, and the processor receives an image from a user, preprocesses the image, converts the preprocessed image into a High Dynamic Range (HDR) image using artificial intelligence, reduces the HDR image, and inputs the reduced HDR image to a pre-trained model to convert it into a high-resolution image.

[0017] According to another embodiment of the present specification, a recording medium according to the present specification is a non-volatile recording medium readable by a computer, in which a program for a method executable by a processor of an image conversion apparatus is recorded, and the method includes: a first step of determining whether data input from a user is an image or a video; a second step of preprocessing an image if the data input from the user is an image; a third step of converting the preprocessed image into a High Dynamic Range (HDR) image using artificial intelligence; a fourth step of reducing the HDR image; and a fifth step of inputting the reduced HDR image to a pre-trained model and converting it into a high-resolution image.

[0018] As described above, according to the present specification, by providing method and apparatus for converting images based on artificial intelligence that preprocesses an image input from a user, converts the preprocessed image into a High Dynamic Range (HDR) image using artificial intelligence, reduces the converted HDR image, and then inputs the reduced HDR image to a pre-trained model to convert it into a high-resolution image, it is possible to provide an omnidirectional platform service that can be used not only by specific camera vendors but also by any camera vendor.

[0019] In addition, by providing method and apparatus for converting images based on artificial intelligence that converts an image through HDR (High Dynamic Range), an algorithm that utilizes the brightest and darkest parts of a photograph, it is possible to give a stark contrast to the image.

[0020] In addition, by providing method and apparatus for converting images based on artificial intelligence that removes noise and makes up frame gaps through frame interpolation technology, it is possible to eliminate flickering that occurs when there is a large difference between the previous / next frames of video data.

[0021] FIG. 1 is a block diagram of an image conversion apparatus according to an embodiment of the present invention.

[0022] FIG. 2 is a block diagram illustrating in detail a configuration of the image conversion apparatus according to an embodiment of the present invention.

[0023] FIG. 3 is a block diagram including various modules for converting an image based on artificial intelligence according to an embodiment of the present invention.

[0024] FIG. 4 is a flow chart illustrating an image conversion method based on artificial intelligence according to an embodiment of the present invention.

[0025] It should be noted that the technical terms used in the present specification are used only to describe specific embodiments and are not intended to limit the present invention. In addition, the technical terms used in the present specification should be interpreted as meanings commonly understood by those having ordinary knowledge in the technical field to which the present invention pertains, unless specifically defined otherwise in the present specification, and should not be interpreted in an excessively broad or excessively reduced meaning. In addition, when a technical term used in the present specification is an incorrect technical term that does not accurately express the concept of the present invention, it should be understood as a technical term that can be correctly understood by those skilled in the art. In addition, general terms used in the present invention should be interpreted according to their definitions in dictionaries or according to the context and should not be interpreted in an excessively reduced meaning.

[0026] In addition, the singular expressions used in the present specification include plural expressions unless the context clearly indicates otherwise. In this application, terms such as "configured" or "includes" should not be construed as necessarily including all of the various components or steps described in the specification, and some of the components or steps may not be included, or additional components or steps may be further included.

[0027] In addition, the suffixes "module" and "unit" for components used in the present specification are given or used interchangeably only for the ease of writing the specification, and do not have distinct meanings or roles in themselves.

[0028] In addition, terms including ordinal numbers such as "first" and "second" used in the present specification may be used to describe various components, but the components should not be limited by the terms. The terms are only 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 referred to as the second component, and similarly, the second component may also be referred to as the first component.

[0029] In addition, the model used in the present specification may be constructed in consideration of the applicable field of the model, the computer performance of the device, and the like. As an example, the trained model may be set to output a high-resolution image using a low-resolution image as input data. The trained model may be, for example, a model based on a neural network. The model may be designed to simulate the structure of the human brain on a computer and may include a plurality of network nodes having weights that simulate neurons in the human neural network. The plurality of network nodes may each form a connection relationship to simulate the synaptic activity of neurons exchanging signals through synapses. In addition, the model may include, for example, a neural network model or a deep learning model developed from the neural network model. In the deep learning model, the plurality of network nodes may be located at different depths (or layers) and exchange data according to a convolution connection relationship. Examples of the model may include, but are not limited to, Deep Neural Network (DNN), Recurrent Neural Network (RNN), and Bidirectional Recurrent Deep Neural Network (BRDNN).

[0030] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that the same or similar components will be given the same reference numerals regardless of the drawing reference numerals, and redundant descriptions thereof will be omitted.

[0031] In addition, in describing the present invention, detailed descriptions of related known technologies will be omitted if it is determined that they may obscure the gist of the present invention. In addition, it should be noted that the accompanying drawings are only for facilitating easy understanding of the concept of the present invention, and the concept of the present invention should not be construed as being limited by the accompanying drawings.

[0032] FIG. 1 illustrates a block diagram of an image conversion apparatus according to an embodiment of the present invention. As shown in FIG. 1, the image conversion apparatus 100 includes a display 110, a memory 120, a user input unit 130, and a processor 140. The components shown in FIG. 1 are exemplary diagrams for implementing the embodiments of the present disclosure, and appropriate hardware / software configurations that are obvious to those skilled in the art may be additionally included in the image conversion apparatus 100.

[0033] The display 110 may provide various screens. In particular, the display 110 may display an image or a video composed of a plurality of frames.

[0034] The memory 120 may store instructions or data related to at least one other component of the image conversion apparatus 100. In particular, the memory 120 may be implemented as a non-volatile memory, a volatile memory, a flash memory, a hard disk drive (HDD), or a solid state drive (SSD), and the like. The memory 120 is accessed by the processor 140, and reading / writing / modification / deletion / updating of data by the processor 140 may be performed. In this disclosure, the term memory may include the memory 120, a ROM (not shown), a RAM (not shown) in the processor 140, or a memory card (not shown) mounted on the electronic device 100 (for example, a micro SD card or a memory stick). In addition, programs and data for configuring various screens to be displayed on the display area of the display unit 110 may be stored in the memory 120.

[0035] In addition, the memory 120 may store training data necessary for training the model and may store the model of the present disclosure.

[0036] In addition, as shown in FIG. 3, the memory 120 may store a frame separation module 121, a preprocessing module 122, a High Dynamic Range (HDR) conversion module 123, an auto-contrast module 124, a super-resolution module 125, and a frame merging module 126.

[0037] The frame separation module 121 generates a frame set by splitting a video into a plurality of frames when the data input from the user is a video.

[0038] The preprocessing module 122 preprocesses the image. Here, the preprocessing may include linearization for linearizing the image, normalization for adjusting the contrast and brightness of the entire image, demosaicing for restoring color information, denoising for removing noise and improving sharpness, white balancing for adjusting the color temperature to adjust the color balance of the image, and gamma correction for adjusting the brightness of the image.

[0039] The HDR conversion module 123 converts the preprocessed image (that is, the LDR image) into an HDR image using artificial intelligence. That is, the HDR conversion module 123 according to the present invention recognizes the contrast in the image and widens the range thereof. In other words, the HDR conversion module 123 divides the bright part more precisely to make it brighter and divides the dark part more finely to make it darker.

[0040] To this end, the HDR conversion module 123 according to the present invention may convert the preprocessed image into an HDR image in two ways.

[0041] In the first method, a neural network is configured by arranging a very large number of convolutional layers for inverse tone mapping for HDR image restoration. Specifically, in this method, the upper grayscale and the lower grayscale are divided for the input image signal, and then the original input signal is used for the image for the lower grayscale area, and the image signal of the upper grayscale is inferred through a Convolutional Neural Network (CNN), and then the two signals are appropriately combined to generate the final HDR image.

[0042] In the second method, for image output of a multi-exposure stack, images having various exposure levels with the same dynamic range are generated based on the input image signal, and then they are combined to generate the final HDR image. Here, the multi-exposure stack means a stack composed of LDR images with the same dynamic range having various exposure levels.

[0043] The auto-contrast module 124 reduces the HDR image. That is, the auto-contrast module 124 may reduce the HDR image by emphasizing the contrast of the HDR image (auto-contrast) and trimming the size and rotation of the HDR image (resizing and fix orientation). Specifically, the auto-contrast module 124 may analyze the pixel value distribution of the image to identify the pixel value ranges of the dark part and the bright part, set the minimum value and the maximum value of the image based on the analyzed pixel value ranges, and adjust the contrast according to the set minimum value and maximum value.

[0044] The super-resolution module 125 inputs the reduced HDR image to the pre-trained model and converts it into a high-resolution image. To this end, the super-resolution module 125 may train the model 210 using the reduced HDR image and the original image as a dataset, as shown in FIG. 3.

[0045] The super-resolution module 125 may be configured to receive the reduced HDR image (Low Resolution, LR) from the auto-contrast module 124 and perform super-resolution, that, is, increase the resolution. Hereinafter, the result of the super-resolution will be referred to as a 'high-resolution image.'

[0046] Here, the super-resolution module 125 may be pre-trained to output a high-resolution image when a low-resolution image is input. Hereinafter, the process of training the super-resolution module 125 will be described in more detail.

[0047] First, training data for training the super-resolution module 125 is prepared. The training data may be, for example, a well-known DIV2K dataset. The DIV2K dataset may include a plurality of original high-resolution images.

[0048] Although not shown in FIG. 3, each of the plurality of original high-resolution images included in the DIV2K dataset is downsampled by a separate downsampling module. At this time, according to an embodiment, each of the plurality of original high-resolution images may pass through an anti-aliasing low-pass filter. After that, each of the plurality of original high-resolution images thus downsampled may be compressed by a separate compression module according to various different compression rates. As a result, a plurality of low-resolution images compressed at different compression rates may be prepared.

[0049] After that, the super-resolution module 125 may be trained by a separate training module, using the plurality of original high-resolution images prepared as described above as correct answers and the plurality of low-resolution images compressed at different compression rates as inputs. Here, 'training' may be machine learning, such as deep learning, but is not limited thereto.

[0050] At this time, the low-resolution image used for training may be compressed at different compression rates as described above.

[0051] In addition, in the training process, training may be performed on the super-resolution module 125 for 300 epochs using a low-resolution image compressed at 2.0 bpp and the corresponding original high-resolution image, and when the training is completed, training may be performed on the super-resolution module 125 for 150 epochs using a low-resolution image compressed at another value of bpp and the corresponding high-resolution image. Here, the compression rate for the low-resolution image used when training is performed and the value of epoch mentioned in each are merely exemplary.

[0052] Meanwhile, the super-resolution module 125 described above may include a plurality of residual block units. At this time, each of the plurality of residual block units may be configured to include a convolution layer and a ReLU layer but may be configured not to include a batch normalization layer. By configuring each of the plurality of residual block units in this way, the memory footprint can be smaller and the performance can be further improved.

[0053] When data exists at the same location in a plurality of frames, the frame merging module 126 takes the average value of the data at the same location in all frames and applies it to the corresponding location in the merged frame, and when data does not exist in one or more frames, it takes the average value of the data at the same location in the remaining frames excluding the frames where data does not exist and applies it to the corresponding location in the merged frame, thereby generating a merged frame. Alternatively, the frame merging module 126 selects the corresponding location in the merged frame as an occluded interval when data does not exist in all frames.

[0054] As described above, many occluded intervals included in each frame may be interpolated in the process of merging a plurality of frames. However, for intervals where data does not exist in all frames, the corresponding interval in the merged frame also remains as an occluded interval. For such occluded intervals, the frame merging module 126 can interpolate the video using an in-screen interpolation method or prediction of motion vectors through re-search to generate a final interpolated video.

[0055] Referring back to FIG. 1, the user input unit 130 can receive various user inputs and transmit them to the processor 140. Specifically, the user input unit 130 may include a touch sensor, a (digital) pen sensor, a pressure sensor, or a key. The touch sensor may use, for example, at least one of a capacitive, pressure-sensitive, infrared, or ultrasonic method. The (digital) pen sensor may be, for example, a part of the touch panel or may include a separate recognition sheet. The key may include, for example, a physical button, an optical key, or a keypad.

[0056] In particular, the user input unit 130 can acquire an input signal according to user input of touching a specific image after pressing a specific button (for example, mouse input or keyboard input) or a preset user touch (for example, a long-press touch) to input an image or video. The user input unit 130 can transmit the input signal to the processor 140.

[0057] The processor 140 can be electrically connected to the display 110, the memory 120, and the user input unit 130 to control the overall operation and functions of the image conversion apparatus 100. In particular, the processor 140 can perform image conversion functions related to the image input by the user using the frame separation module 121, the preprocessing module 122, the HDR conversion module 123, the auto-contrast module 124, the super-resolution module 125, and the frame merging module 126. In particular, the processor 140 can acquire an input signal according to user input for inputting an image or video using the user input unit 130, acquire the image or video in response to the input signal, control the display 110 to display the image or video, and convert the image (or each of the plurality of frames included in the video) into a high-resolution image.

[0058] FIG. 2 is a block diagram illustrating in detail a configuration of the image conversion apparatus according to an embodiment of the present invention. As shown in FIG. 2, the image conversion apparatus 100 may include a display 110, a memory 120, a user input unit 130, a processor 140, a camera 150, a communication unit 160, and an audio output unit 170. Meanwhile, the display 110, the memory 120, and the user input unit 130 have been described with reference to FIG. 1, and thus redundant descriptions thereof will be omitted.

[0059] The camera 150 can capture videos and images. At this time, the camera 150 may be provided on at least one of the front and rear of the image conversion apparatus 100. Meanwhile, the camera 150 may be provided inside the image conversion apparatus 100, but this is only an embodiment, and the camera 150 may exist outside the image conversion apparatus 100 and be connected to the image conversion apparatus 100 wirelessly or wiredly.

[0060] The communication unit 160 can perform communication with various types of external devices according to various types of communication methods. The communication unit 160 may include at least one of a Wi-Fi chip 161, a Bluetooth chip 162, a wireless communication chip 163, and an NFC chip 164. The processor 140 can perform communication with an external server or various external devices using the communication unit 160.

[0061] In particular, the communication unit 160 can perform communication with an external detection server or an external cloud server.

[0062] The audio output unit 170 is configured to output various audio data that have undergone various processing operations such as decoding, amplification, and noise filtering by an audio processing unit (not shown), as well as various notification sounds or voice messages. In particular, the audio output unit 170 may be implemented as a speaker, but this is only an embodiment, and the audio output unit 170 may be implemented as an output terminal capable of outputting audio data.

[0063] In particular, the audio output unit 170 can provide information on the conversion result to the user in audio form.

[0064] The processor 140 (or the control unit) can control the overall operation of the image conversion apparatus 100 using various programs stored in the memory 120.

[0065] The processor 140 may include a RAM 141, a ROM 142, a graphic processing unit 143, a main CPU 144, first to n-th interfaces 145-1 to 145-n, and a bus 146. At this time, the RAM 141, the ROM 142, the graphic processing unit 143, the main CPU 144, and the first to n-th interfaces 145-1 to 145-n may be connected to each other through the bus 146.

[0066] FIG. 4 is a flow chart illustrating an image conversion method based on artificial intelligence according to an embodiment of the present invention.

[0067] Referring to FIG. 4, the image conversion apparatus 100 receives a file uploaded from a user (S410).

[0068] The image conversion apparatus 100 determines whether the uploaded file is an image or a video (S420), and if the uploaded file is a video, divides the video into a plurality of frames (S422).

[0069] The image conversion apparatus 100 preprocesses the image if the uploaded file is an image (S430). The image conversion apparatus 100 preprocesses each frame if the uploaded file is a video. Here, the preprocessing may include linearization for linearizing the image, normalization for adjusting the contrast and brightness of the entire image, demosaicing for restoring color information, denoising for removing noise and improving sharpness, white balancing for adjusting the color temperature to adjust the color balance of the image, and gamma correction for adjusting the brightness of the image.

[0070] The image conversion apparatus 100 converts the preprocessed image (that is, the LDR image) into an HDR image using artificial intelligence (S440). That is, the image conversion apparatus 100 recognizes the contrast in the image and widens the range thereof. In other words, the image conversion apparatus 100 divides the bright part more precisely to make it brighter and divides the dark part more finely to make it darker.

[0071] At this time, the image conversion apparatus 100 may convert the preprocessed image into an HDR image in two ways.

[0072] In the first method, a neural network is configured by arranging a very large number of convolutional layers for inverse tone mapping for HDR image restoration. Specifically, in this method, the upper grayscale and the lower grayscale are divided for the input image signal, and then the original input signal is used for the image for the lower grayscale area, and the image signal of the upper grayscale is inferred through a Convolutional Neural Network (CNN), and then the two signals are appropriately combined to generate the final HDR image.

[0073] In the second method, for image output of a multi-exposure stack, images having various exposure levels with the same dynamic range are generated based on the input image signal, and then they are combined to generate the final HDR image. Here, the multi-exposure stack means a stack composed of LDR image with the same dynamic range having various exposure levels.

[0074] Next, the image conversion apparatus 100 reduces the HDR image (S450). At this time, the image conversion apparatus 100 may reduce the HDR image by emphasizing the contrast of the HDR image (auto-contrast) and trimming the size and rotation of the HDR image (resizing and fix orientation). Specifically, the image conversion apparatus 100 may analyze the pixel value distribution of the image to identify the pixel value ranges of the dark part and the bright part, set the minimum value and the maximum value of the image based on the analyzed pixel value ranges, and adjust the contrast according to the set minimum value and maximum value.

[0075] The image conversion apparatus 100 inputs the reduced HDR image to the pre-trained model and converts it into a high-resolution image (S460). To this end, the image conversion apparatus 100 may train the model 210 using the reduced HDR image and the original image as a dataset.

[0076] Finally, the image conversion apparatus 100 merges the plurality of frames converted through the steps of S430 to S460 as described above to generate a video if the uploaded file is a video (S470). At this time, when data exists at the same location in the plurality of frames, the image conversion apparatus 100 takes the average value of the data at the same location in all frames and applies it to the corresponding location in the merged frame, and when data does not exist in one or more frames, it takes the average value for the data at the same location in the remaining frames excluding the frames where data does not exist and applies it to the corresponding location in the merged frame, thereby generating a merged frame. Alternatively, the image conversion apparatus 100 selects the corresponding location in the merged frame as an occluded interval when data does not exist in all frames.

[0077] As described above, many occluded intervals included in each frame may be interpolated in the process of merging a plurality of frames. However, for intervals where data does not exist in all frames, the corresponding interval in the merged frame also remains as an occluded interval. The method may further include a step of interpolating the video for such occluded intervals using an in-screen interpolation method or prediction of motion vectors through re-search to generate a final interpolated video.

[0078] Various embodiments of the present disclosure may be implemented by software including instructions stored on a machine-readable storage medium (for example, a computer). The apparatus may include an electronic device (for example, the automatic object detection apparatus 100) according to the disclosed embodiments, which is an apparatus capable of calling stored instructions from a storage medium and operating according to the called instructions. When the instructions are executed by the processor, the processor may directly perform the function corresponding to the instructions or may perform the function using other components under the control of the processor. The instructions may include code generated or executed by a compiler or an interpreter. The machine-readable storage medium may be provided in the form of a non-transitory storage medium. Here, 'non-transitory' means that the storage medium does not include a signal and is tangible but does not distinguish between data being semi-permanently or temporarily stored on the storage medium.

[0079] According to an embodiment, the method according to various embodiments disclosed in this document may be provided included in a computer program product. The computer program product may be traded between a seller and a buyer as a commodity. The computer program product may be distributed in the form of a machine-readable storage medium (for example, a compact disc read only memory (CD-ROM)), or online through an application store (for example, PlayStoreTM).Inthecaseofonlinedistribution,atleastpartofthecomputerprogramproductmaybeatleasttemporarilystoredonastoragemediumsuchasamemoryofarelayserver,aserveroftheapplicationstore,orthemanufacturer'sserver,ormaybetemporarilygenerated.

[0080] Each component (for example, a module or a program) according to various embodiments may be configured by a single or a plurality of entities, some of the sub-components among the sub-components as described above may be omitted, or other sub-components may be further included in various embodiments. Alternatively or additionally, some components (for example, modules or programs) may be integrated into one entity to perform the same or similar functions performed by the respective components before being integrated. Operations performed by modules, programs, or other components according to various embodiments may be executed sequentially, in parallel, repeatedly, or heuristically, at least some operations may be executed in a different order, omitted, or other operations may be added.

[0081] [Reference Numerals]

[0082] 121: Frame Separation Module

[0083] 122: Preprocessing Module

[0084] 123: HDR Conversion Module

[0085] 124: Auto-contrast Module

[0086] 125: Super-resolution Module

[0087] 126: Frame Merging Module

Claims

1.An image conversion method of an image conversion apparatus based on artificial intelligence, the method comprising:a first step of determining whether data input from a user is an image or a video;a second step of preprocessing an image if the data input from the user is an image;a third step of converting the preprocessed image into a High Dynamic Range (HDR) image using artificial intelligence;a fourth step of reducing the HDR image; anda fifth step of inputting the reduced HDR image to a pre-trained model and converting it into a high-resolution image.2.The image conversion method based on artificial intelligence of Claim 1, the method further comprising:a first sub-step of the first step of dividing a video into a plurality of frames if the data input from the user is a video; anda sixth step of performing the second to fifth steps on each frame, and then merging the converted plurality of frames to generate a video.3.The image conversion method based on artificial intelligence of Claim 2, the method further comprising:a first sub-step of the fifth step of performing frame interpolation between two consecutive frames.4.The image conversion method based on artificial intelligence of Claim 1, wherein the preprocessing includes at least one of linearization, normalization, demosaicing, denoising, white balancing, and gamma correction.5.The image conversion method based on artificial intelligence of Claim 1, wherein the fourth step includes emphasizing contrast of the HDR image and trimming size and rotation of the HDR image.6.The image conversion method based on artificial intelligence of Claim 1, wherein the third step includes converting the preprocessed image into the HDR image through inverse tone mapping.7.The image conversion method based on artificial intelligence of Claim 1, wherein the fifth step includes training the model using the reduced HDR image and an original image as a dataset.8.An image conversion apparatus, comprising:a memory storing one or more instructions; anda processor executing the one or more instructions,wherein the processor receives an image from a user, preprocesses the image, converts the preprocessed image into a High Dynamic Range (HDR) image using artificial intelligence, reduces the HDR image, and inputs the reduced HDR image to a pre-trained model to convert it into a high-resolution image.9.A non-volatile computer-readable recording medium having recorded thereon a program for a method executable by a processor of an image conversion apparatus, the method comprising:a first step of determining whether data input from a user is an image or a video;a second step of preprocessing an image if the data input from the user is an image;a third step of converting the preprocessed image into a High Dynamic Range (HDR) image using artificial intelligence;a fourth step of reducing the HDR image; anda fifth step of inputting the reduced HDR image to a pre-trained model and converting it into a high-resolution image.

Images