KR102154424B1 - Advanced system and method for video compression - Google Patents

Abstract

The present invention relates to an improved image compression system and method. In the disclosed improved image compression system and method, an error pattern of a distorted image that occurs in the process of converting an analog image into a digital form for a computer and compressing it is identified and classified through machine learning, and then the distorted image is restored. In the process of decoding a digitally coded compressed image and converting it into an analog signal corresponding thereto, an encoder that generates a compensable compensation value is transmitted from the encoder through a bitstream and restored. It provides an improved video compression system and method comprising a decoder. According to the present invention, an improved image compression system that uses fewer bits when compressing an image and transmits information about the original image to a decoder, thereby improving image compression efficiency and improving image quality compared to the prior art. And there is an advantage of being able to provide a method.

Description

Improved video compression system and method {ADVANCED SYSTEM AND METHOD FOR VIDEO COMPRESSION}

The present invention relates to an improved image compression system and method, and more particularly, an improved image compression system that classifies an error of a distorted image using a machine learning-based method and finds an appropriate compensation value to improve compression performance. And a method.

The content described in this section merely provides background information on the present embodiment and does not constitute the prior art.

Along with the rapidly developing information technology society, video/video is also being used in various ways in real life. As images/movies are closely related to the daily life of general consumers, various digital images are acquired and used. As videos are closely related to the daily life of citizens, videos use most of the internet traffic, and the degree is increasing. When a video is first created and saved, compression inevitably proceeds.

Image compression is an operation of reducing a vast amount of image data. Compressed video effectively reduces the bandwidth required to deliver digital video through cable, terrestrial broadcasting, and satellite services. The image compression method is classified into lossy compression and lossless compression depending on whether or not all information of the original image is retained when restoring after image compression. Lossless compression is a compression method in which the compressed image is completely consistent with the original image by completely coating and decoating the image, whereas lossy compression is a compression method by removing redundant or less important information from the image. It is a compression method that increases Most compressions use lossy compression to reduce capacity, and in general, joint picture expert groups (JPEG) and motion pic-ture expert groups (MPEG) are standard compression techniques for still images and moving pictures.

As display devices are gradually developed, the size of images mainly used by people is increasing, whereas bandwidth is limited in order to share a moving image through the Internet, so the smaller the amount of bits for compression, the more useful.

An important consideration to consider as the utilization of video increases is that the video quality must be high while using a small amount of bits. In other words, it is necessary to increase the compression efficiency of the video.

Accordingly, in the present invention, an image compression system and method with improved compression efficiency compared to the prior art are proposed.

Korean Patent Laid-Open Publication No. 10-2018-0119753, November 5, 2018 (Name: Machine learning algorithm for image restoration using compression parameters and image restoration method using the same) Korean Laid-Open Patent Publication No. 10-2018-0100976, published on September 12, 2018 (Name: image encoding/decoding method and apparatus using blur image learning based on deep neural networks) Korean Registered Patent Publication No. 10-1885855, published on August 7, 2018 (Name: image signal transmission using high resolution estimation technique)

(Non-Patent Document 1) Park, Woon-Sung, and Muncurl Kim. "CNN-based in-loop filtering for coding efficiency improvement" Image, Video and Multidimensional Signal Processing Workshop (IVMSP). 2016 IEEE 12th IEEE. 2016 (Non-Patent Document 2) Dai Yuanying Dong Liu, and Feng Wu "A convolutional neural network approach for post-processing in HEVC Intra coding "International Conference on Multimedia Modeling. Springer, Cham, 2017.

The present invention has been proposed to solve the problems of the prior art described above, and provides an improved image compression system and method for improving image compression efficiency compared to the prior art by using fewer bits when compressing an image. It has a main purpose.

In addition, another object of the present invention is to provide an improved image compression system and method capable of improving image quality by sending information about an original image to a decoder while increasing image compression efficiency through fewer bits.

The problem to be solved of the present invention is not limited to those mentioned above, and another problem to be solved that is not mentioned will be clearly understood by those of ordinary skill in the art from the following description.

One aspect of the present invention for achieving the above object is to grasp and classify an error pattern of a distorted image generated in the process of compressing an analog image into a digital form for a computer through machine learning, and then, the distortion An encoder that generates a compensation value capable of reconstructing the image; And a decoder that receives and restores a compensation value of the compressed image from the encoder through a bitstream in the process of decoding the digitally coded compressed image and converting it into an analog signal corresponding thereto. Provides an image compression system.

Another aspect of the present invention is an improved image compression encoder for improving image compression efficiency using machine learning, receiving a distorted image as an input, identifying and classifying the characteristics of errors included in the distorted image An error classification module; And a residual image compression module that receives the original image and the distorted image as inputs and finds a compensation value capable of compensating for the error classified by the error classification module. to provide.

Another aspect of the present invention is an improved image compression decoder for improving image compression efficiency using machine learning, in which a distorted image is received as an input, and the characteristics of errors included in the distorted image are identified and classified. An error classification module; And an image restoration module for restoring a damaged image by receiving a compensation value for restoration of the distorted image from an encoder through a bitstream.

Another aspect of the present invention is an improved image compression encoding method for improving image compression efficiency using machine learning, in which an error characteristic is determined according to a pattern classification method previously learned by inputting a distorted image from an encoder. Extracting; Classifying a pattern of the extracted error features; Extracting and compressing residual images for differences between the distorted and original images as inputs; Classifying a pattern of the extracted and compressed residual image; And generating an appropriate compensation value for reconstructing the distorted image according to the classified error pattern of the residual image.

Another aspect of the present invention provides a computer-readable recording medium recording a program for executing the improved video compression encoding method.

Another aspect of the present invention is an improved image compression decoding method for improving image compression efficiency using machine learning, receiving a distorted image as an input and grasping the characteristics of errors included in the distorted image Extracting; Classifying a pattern of the extracted error features; Receiving a compensation value for reconstructing the distorted image from an encoder through a bitstream; And applying the received compensation value to the distorted image to restore an image.

Another aspect of the present invention provides a computer-readable recording medium recording a program for executing the improved video compression decoding method.

According to the improved image compression system and method of the present invention, there is an effect of providing an improved image compression system and method that improves image compression efficiency compared to the prior art by using fewer bits when compressing an image. have.

In addition, there is an effect that it is possible to provide an improved image compression system and method capable of increasing image quality by sending information about an original image to a decoder while increasing image compression efficiency through fewer bits.

The effects that can be obtained in the present invention are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those of ordinary skill in the art from the following description. .

BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are included as part of the detailed description to aid in understanding of the present invention, provide embodiments of the present invention, and together with the detailed description, the technical features of the present invention will be described.
1 is a diagram illustrating a configuration of an improved image compression system according to an embodiment of the present invention.
2 is a diagram illustrating a configuration of an error classification module according to an embodiment of the present invention.
3 is a diagram illustrating a configuration of a residual image compression module according to an embodiment of the present invention.
4 is a diagram illustrating an improved image compression 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. The detailed description to be disclosed hereinafter together with the accompanying drawings is intended to describe exemplary embodiments of the present invention, and is not intended to represent the only embodiments in which the present invention may be practiced. The following detailed description includes specific details to provide a thorough understanding of the present invention. However, those of ordinary skill in the art to which the present invention pertains knows that the present invention may be practiced without these specific details.

In some cases, in order to avoid obscuring the concept of the present invention, well-known structures and devices may be omitted, or may be shown in a block diagram form centering on core functions of each structure and device.

Throughout the specification, when a part is said to "comprising or including" a certain component, it means that other components may be further included rather than excluding other components unless specifically stated to the contrary. do. In addition, terms such as "... unit", "... group", and "module" described in the specification mean units that process at least one function or operation, which can be implemented by hardware or software or a combination of hardware and software. have. In addition, "a or an", "one", "the" and similar related words are different from this specification in the context of describing the invention (especially in the context of the following claims). Unless indicated or clearly contradicted by context, it may be used in the sense of including both the singular and the plural.

In describing the embodiments of the present invention, if it is determined that a detailed description of a known function or configuration may unnecessarily obscure the subject matter of the present invention, a detailed description thereof will be omitted. In addition, terms to be described later are terms defined in consideration of functions in an embodiment of the present invention, which may vary according to the intention or custom of users or operators. Therefore, the definition should be made based on the contents throughout this specification.

Each of the components in the drawings of the present invention is shown independently to represent different characteristic functions in the improved image classification system and method, and does not mean that each component is made up of separate hardware or one software component unit. Does not. That is, each constituent part is listed and included as a constituent part for convenience of explanation, and at least two of the constituent parts are combined to form a single constituent part, or one constituent part is divided into a plurality of constituent parts to perform a function Integrated embodiments and separate embodiments of the components are also included in the scope of the present invention unless departing from the essence of the present invention.

In addition, some of the components are not essential components that perform essential functions in the present invention, but may be optional components only for improving performance. The present invention can be implemented by including only the components essential to implement the essence of the present invention excluding components used for performance improvement, and a structure including only essential components excluding optional components used for performance improvement Also included in the scope of the present invention.

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

1 is a diagram illustrating a configuration of an improved image compression system according to an embodiment of the present invention.

The encoder 100 refers to a device or program for converting a video or audio signal into another signal format. It is mainly used to convert an analog image into a digital form for a computer.

In an embodiment of the present invention, the encoder 100 includes an error classification module 110 and a residual image compression module 120.

The error classification module 110 is a module that receives a distorted image as an input and uses a machine learning algorithm to identify and classify characteristics of an error.

Machine learning algorithms include Decision Tree, Bayesian network, Support Vector Machine (SVM), artificial neural network, and deep learning. The support vector machine model is a statistical learning theory that is in the spotlight in pattern classification, and is mainly used for classification and regression analysis as a model for supervised (supervised/teacher) learning for pattern recognition and data analysis. In other words, after estimating a decision function using the probability distribution obtained in the learning process for the learning diagnosis of learning data and category information, new data is classified in two ways according to this function. Supervised learning is a method of first teaching information to a computer, and the computer classifies information based on the result of learning in advance. Deep learning is a machine learning technology built on the basis of an artificial neural network (ANN) to enable computers to learn on their own like a human using multiple data. The human brain discovers patterns in numerous data. After that, the computer learns the machine to discern things by imitating the information processing method that separates things. When deep learning technology is applied, computers can recognize, infer, and make decisions on their own, even if humans do not set all judgment criteria.

A moving picture is a set of still pictures that are sequentially displayed over a short time interval, and a very high correlation exists between neighboring images due to a short time interval between these still pictures. Compression in the same picture as in JPEG is called intra-mode encoding, and a picture compressed by the intra-picture encoding method is called I(Intra)-frame.

Encoding using a predictive compression method with reference to a frame in the past or future is referred to as inter mode encoding. The difference between the previous screen and the current screen is very small, and only the difference value is encoded. For inter-screen encoding, the concept of motion estimation and motion compensation in units of macroblocks is used. There are P (Predictive)-frame and B (Bidirectional)-frame in pictures compressed by the inter-picture coding method.

In order to compress video with high efficiency, redundancy is removed by using various techniques such as motion prediction and motion compensation. However, as the compression rate increases, the quality of the image deteriorates, and an image distorted compared to the original image is generated.

The error classification module 110 may classify the types of errors according to a pre-learned pattern classification method using an input image and a distorted image given from an encoder. The error referred to at this time is the difference between the original image and the distorted image. When classifying an error, the error classification module 110 outputs the number of channels corresponding to the number according to the number of gadgets to be classified. At this time, the output channels can be regarded as classified errors.

The error classification module 110 is a device to which a technology for classifying the type of error by analyzing the characteristics of a distorted image given by a user based on artificial intelligence is applied.

It is preferable that the error classification module 110 uses a deep learning convolutional neural network (CNN) technique. Convolutional neural networks (CNNs) are a type of multilayer perceptrons designed to use minimal preprocessing. A convolutional neural network (CNN) is composed of one or several convolutional layers and a general artificial neural network layer on top of it, and additionally utilizes weights and pooling layers. Through this structure, the convolutional neural network (CNN) can sufficiently utilize the input data of a two-dimensional structure, and can bring good performance in both video and audio fields compared to other deep learning structures.

The residual image compression module 120 receives the original image and the distorted image as inputs and finds a compensation value that can best compensate for the error classified by the error classification module 110. When the error classified by the error classification module 110 is determined, the compensation value in the residual image compression module 120 is determined as a difference between the actual value and the error value.

When the error classification module 110 classifies the errors according to the criteria, the residual image compression module 120 may find and restore a compensation value suitable for the characteristics of each classified error. In order to find such a compensation value, the residual image compression module 120 receives the distorted image and the original image as inputs, compresses the residual image made of them, and classifies the error according to the same pattern classification method as the error classification module 110. It can be classified into several categories, and it can be used as a reward value. Accordingly, the error classified by the error classification module 110 may be compensated by using the compensation value output from the residual image compression module 120 to restore a damaged image.

The error classification module 110 and the residual image compression module 120 may interoperate with each other to provide feedback and perform learning more effectively. When an error is generated in a given network learning process, this error is designed to simultaneously affect the error classification module 110 and the residual image compression module 120, so that when the two modules are trained, they are learned by referring to each other.

The decoder 200 refers to a device or program that decodes digitally coded data by the encoder 100 and converts it into an analog signal corresponding thereto.

In an embodiment of the present invention, the decoder 200 includes an error classification module 110.

Since the original image can be used when encoding in the encoder 100, the residual image can be classified and compressed in the residual image compression module 120 of the artificial neural network, but the original image cannot be used when decoding in the decoder 200. Therefore, the residual image cannot be used.

Accordingly, the decoder 200 only uses the error classification module 110, but additionally, the error compensation value is transmitted from the encoder 100 through the bitstream, and the image restoration module 210 restores the damaged image.

As a result of an experiment applying the improved image compression system according to an embodiment of the present invention, it was confirmed that the performance was improved by 5% for HD-sized images and 13% for UHD-level images.

Specifically, among the three methods used for video compression (All Intra, Low Delay P, Random Access), it was conducted in a Low Delay P environment, and experiments were conducted on 20 test images based on 45 learning images. As a result, the results are different depending on the image size, with 13% bit reduction in UHD size, 9% in 2560x1600 size, and 5% performance improvement in HD size.

2 is a diagram illustrating a configuration of an error classification module according to an embodiment of the present invention.

The error classification module 110 may be divided into a feature extraction unit 111 and a pattern classification unit 112.

The feature extracting unit 111 has a shape of a convolutional neural network (CNN) attaching a feature extracting unit 111, which is a convolutional layer of several layers, in front of the pattern classifying unit 112, which is a neural network. The convolutional neural network (CNN) extracts the features of the error compared to the original image for the distorted image received through the previous convolutional layer, and based on the extracted features, a feature extraction unit ( 111).

The feature extracting unit 111 is composed of a filter that functions to extract features again, and an activation function that converts a value obtained by applying the filter into a nonlinear value. Since the feature map, which is the result obtained by applying the filter, comes out as a quantitative value, it is necessary to change the filtered feature to a nonlinear value of "yes, no", and this is the activation function.

The feature extraction unit 111 extracts features by repeatedly combining the filter and the activation function.

The pattern classifying unit 112 is composed of a fully connected layer to classify errors in distorted images. For example, information on the edge of an object and information on an end point may be classified separately, or may be classified by the size, location, or shape of the error.

3 is a diagram illustrating a configuration of a residual image compression module according to an embodiment of the present invention.

The residual image compression module 120 may be divided into a residual image extracting unit 121, a residual image classifying unit 122, and a compensation value generating unit 123.

The residual image extracting unit 121 extracts and compresses a residual image of a difference between the distorted image and the original image as inputs.

The residual image classification unit 122 classifies the residual image into a pattern similar to the pattern classification unit 112 of the error classification module 110.

The compensation value generator 123 generates an appropriate compensation value capable of restoring a distorted image according to the pattern classified by the residual image classification unit 122.

4 is a diagram illustrating an improved image compression method according to an embodiment of the present invention.

The type of error may be classified according to the pattern classification method learned in advance in the error classification module 110 by inputting the distorted image to the encoder 100.

First, in the feature extraction unit 111 of the error classification module 110, the convolutional neural network (CNN) is the convolutional neural network (CNN) is the error feature for the distorted image received through the convolutional layer. Is extracted (S401).

Based on the error characteristics of the distorted image, the pattern classifying unit 112, which is a neural network, is composed of a fully connected layer and classifies the error of the distorted image (S403). For example, information on the edge of an object and information on an end point may be classified separately, or may be classified by the size, location, or shape of the error.

The residual image extraction unit 121 of the residual image compression module 120 receives the distorted image and the original image as inputs, and extracts and compresses the residual image for the difference between these images (S405).

The residual image classification unit 122 classifies a pattern similar to that performed by the pattern classification unit 112 of the error classification module 110 on the residual image extracted and compressed by the residual image extraction unit 121 (S407).

The compensation value generator 123 generates an appropriate compensation value capable of restoring a distorted image according to the pattern of errors classified by the residual image classification unit 122 (S409).

When necessary, the learning data performed by the error classification module 110 and the residual image compression module 120 are interlocked with each other to provide feedback, so that learning can be performed more effectively (S411).

The error classification module 110 of the decoder 200 extracts a feature of an error for the received distorted image, similar to step S401 (S413).

Thereafter, similar to step S403, the features extracted from the error of the distorted image are classified (S415).

In addition, the compensation value generated by the compensation value generator 123 is additionally transmitted to the decoder 200 through a bitstream (S417).

The image restoration module 210 of the decoder 200 restores the damaged image by applying the received compensation value to the distorted image (S419).

In FIG. 4, steps S401 to S419 are described as sequentially executing, but this is only illustrative of the technical idea of the present embodiment, and those of ordinary skill in the technical field to which this embodiment belongs In the range not departing from the essential characteristics, the order shown in FIG. 7 may be changed and executed, or at least one of the steps S401 to S419 may be executed in parallel, and thus various modifications and variations may be applied. It is not limited. For example, the learning data linkage in S411 may be performed before feature extraction and at any time, and the compensation value transmission in S417 may be transmitted before feature extraction in S413 and stored in the decoder 200.

Combinations of each block in the block diagram attached to the present specification and each step in the flowchart may be performed by computer program instructions. Since these computer program instructions can be mounted on a processor of a general purpose computer, special purpose computer, or other programmable data processing equipment, the instructions executed by the processor of the computer or other programmable data processing equipment are shown in each block or flow chart of the block diagram. Each step creates a means to perform the functions described. These computer program instructions can also be stored in computer-usable or computer-readable memory that can be directed to a computer or other programmable data processing equipment to implement a function in a particular way, so that the computer-usable or computer-readable memory It is also possible to produce an article of manufacture in which the instructions stored in the block diagram contain instruction means for performing the functions described in each block or flow chart. Computer program instructions can also be mounted on a computer or other programmable data processing equipment, so a series of operating steps are performed on a computer or other programmable data processing equipment to create a computer-executable process to create a computer or other programmable data processing equipment. It is also possible for the instructions to perform the processing equipment to provide steps for performing the functions described in each block of the block diagram and each step of the flowchart.

In addition, each block or each step may represent a module, segment, or part of code comprising one or more executable instructions for executing the specified logical function(s). In addition, it should be noted that in some alternative embodiments, functions mentioned in blocks or steps may occur out of order. For example, two blocks or steps shown in succession may in fact be performed substantially simultaneously, or the blocks or steps may sometimes be performed in the reverse order depending on the corresponding function.

The above description is merely illustrative of the technical idea of the present invention, and those of ordinary skill in the art to which the present invention pertains will be able to make various modifications and variations without departing from the essential characteristics of the present invention. Accordingly, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention, but to explain the technical idea, and the scope of the technical idea of the present invention is not limited by these embodiments. The scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of the present invention.

According to the improved image compression system and method of the present invention, fewer bits are used when compressing an image, thereby sending information about an original image to a decoder, thereby improving image compression efficiency while improving image quality compared to the prior art. In that it can be used as a solution that can provide an improved image compression system and method that allows it to be used, the possibility of marketing or sales of applied devices is increased as it exceeds the limitations of existing technologies. It is an invention that has industrial applicability because it is not only sufficient, but also to the extent that it can be implemented clearly in reality.

100: encoder
110: error classification module 111: feature extraction unit 112: pattern classification unit
120: residual image compression module 121: residual image extraction unit
122: residual image classification unit 123: compensation value generation unit
200: decoder
210: image restoration module

Claims (19)

An encoder that detects and classifies an error pattern of a distorted image that occurs in the process of converting an analog image into a digital form for a computer and compressing it through machine learning, and then generating a compensation value for restoring the distorted image; And
In the process of decoding the digitally coded compressed image and converting it into an analog signal corresponding thereto, a decoder configured to receive and restore a compensation value of the compressed image from the encoder through a bitstream,
The encoder includes: a residual image extracting unit for extracting and compressing a residual image of a difference between the distorted image and the original image as inputs; A residual image classification unit for classifying the pattern of the residual image; And a compensation value generator for generating a compensation value capable of reconstructing the distorted image according to the classified pattern,
The generated compensation value is transmitted to a decoder to compensate for the distorted image. The method of claim 1,
The generation of the compensation value,
An improved image compression system, characterized in that the type of error is classified according to a previously learned pattern classification method using the input original image and the distorted image given by the encoder. In an improved video compression encoder for improving video compression efficiency using machine learning,
An error classification module for receiving a distorted image as an input and identifying and classifying characteristics of errors included in the distorted image; And
A residual image compression module that takes an original image and the distorted image as inputs and finds a compensation value capable of compensating for the error classified by the error classification module,
The residual image compression module includes: a residual image extracting unit for extracting and compressing a residual image of a difference between the distorted image and the original image as inputs;
A residual image classification unit for classifying the pattern of the residual image; And a compensation value generator for generating a compensation value capable of reconstructing the distorted image according to the classified pattern,
And the generated compensation value is transmitted to a decoder to compensate for the distorted image. The method of claim 3,
The error classification module,
An improved image compression encoder, characterized in that the type of error is classified by analyzing features of a distorted image given by a user based on machine learning. The method of claim 4,
The machine learning,
An improved image compression encoder, characterized in that it is a convolutional neural network (CNN) of deep learning. The method of claim 3,
And the error classification module and the residual image compression module interoperate with each other to provide feedback to share learning data to perform machine learning. The method of claim 3,
The error classification module,
A feature extractor for extracting features of an error compared to the original image for the received distorted image; And
And a pattern classifying unit for classifying an error pattern based on the features extracted from the feature extracting unit. The method of claim 7,
Classification of the error pattern of the distorted image,
An improved image compression encoder, characterized in that at least one of edge information of an object, information of an end point, a size of an error, a location of an error, and a shape of an error. delete delete The method of claim 3,
The residual image is,
The improved image compression encoder, characterized in that the difference calculated by comparing the original image and the distorted image. In an improved image compression decoder for improving image compression efficiency using machine learning,
An error classification module for receiving a distorted image as an input and identifying and classifying characteristics of errors included in the distorted image; And
An image restoration module for restoring a damaged image by receiving a compensation value for restoring the distorted image from an encoder through a bitstream,
The encoder includes: a residual image extracting unit for extracting and compressing a residual image of a difference between the distorted image and the original image as inputs; A residual image classification unit for classifying the pattern of the residual image; And a compensation value generator for generating a compensation value capable of reconstructing the distorted image according to the classified pattern,
And the generated compensation value is transmitted to a decoder to compensate for the distorted image. The method of claim 12,
The error classification module,
An improved image compression decoder, characterized in that by analyzing features of distorted images given by a user based on machine learning, and classifying types of errors. In an improved video compression encoding method for improving video compression efficiency using machine learning,
Extracting an error feature according to a previously learned pattern classification method by inputting the distorted image by an encoder;
Classifying a pattern of the extracted error features;
Extracting and compressing residual images for differences between the distorted and original images as inputs;
Classifying a pattern of the extracted and compressed residual image;
Generating an appropriate compensation value capable of reconstructing the distorted image according to the classified error pattern of the residual image; And
And transmitting the generated compensation value to a decoder to compensate for a distorted image. The method of claim 14,
The pattern of the classified error features,
An improved video compression encoding method, characterized in that at least one of edge information of an object, information of an end point, an error size, an error location, and an error shape. The method of claim 14,
The learning data generated in the step of extracting the features of the error according to the pattern classification method learned in advance by inputting the distorted image from the encoder and classifying the pattern of the extracted and compressed residual image are fed back in conjunction with each other, An improved video compression encoding method, characterized in that improving learning efficiency. A computer-readable recording medium recording a program for executing the improved video compression encoding method according to any one of claims 14 to 16. In an improved image compression decoding method for improving image compression efficiency using machine learning,
Receiving a distorted image as an input, identifying and extracting characteristics of errors included in the distorted image;
Classifying a pattern of the extracted error features;
Receiving a compensation value for reconstructing the distorted image from an encoder through a bitstream; And
And restoring an image by applying the received compensation value to the distorted image,
The compensation value is in the encoder
Extracting and compressing residual images for differences between the distorted and original images as inputs;
Classifying a pattern of the extracted and compressed residual image; And
And generating an appropriate compensation value capable of reconstructing the distorted image according to the classified error pattern of the residual image. A computer-readable recording medium on which a program for executing the improved video compression decoding method according to claim 18 is recorded.

Images