The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
최근에는 디코딩된 영상의 복원 정확도를 향상시키기 위해 유연한 비선형 분석과 합성 변환을 사용하는 컨벌루션 신경망 기반의 영상 압축 시스템이 개발되었습니다. 최적화를 위해 피크 신호 대 잡음비, 다중 스케일 구조 유사성 등의 객관적인 지표를 사용하는 이러한 방법은 높은 객관적인 결과를 얻을 수 있지만 이러한 지표는 인간의 시각적 특성을 반영하지 못하여 주관적인 이미지 품질을 저하시킬 수 있습니다. 주관적 영상 품질을 향상시키는 방법 중 하나로 GAN(Generative Adversarial Network)이라는 프레임워크를 사용하는 방법이 보고된 바 있다. 복원된 이미지의 분포를 자연 이미지에 가깝게 최적화합니다. 따라서 흐릿함, 울림, 차단과 같은 시각적 아티팩트를 억제합니다. 그러나 이러한 방식은 복원된 영상이 주관적으로 자연스러운지 아닌지에 초점을 맞추는 데 최적화되어 있기 때문에 복호화 과정에서 원본 영상과 관련되지 않은 성분들이 복원 영상에 섞여 들어간다. 따라서, 외관이 자연스러워 보이더라도 주관적인 유사성이 저하될 수 있다. 본 논문에서는 기존 GAN 기반 압축 기술이 주관적 유사성을 저하시키는 이유를 조사한 후 서로 다른 확률 분포를 갖는 이미지 소스 간의 GAN 프레임워크에서 이미지 생성을 처리하는 방법을 다시 생각하여 이 문제를 해결했습니다. 본 논문에서는 코딩 특징과 복원된 영상 간의 상호 정보를 최대화하는 방법을 설명합니다. 실험 결과, 제안된 상호 정보량은 주관적 유사도와 명확한 상관관계가 있음을 보여 주며, 이 방법을 사용하면 주관적 유사도가 높은 영상 압축 시스템을 개발할 수 있습니다.
Shinobu KUDO
Nippon Telegraph and Telephone Corporation
Shota ORIHASHI
Nippon Telegraph and Telephone Corporation
Ryuichi TANIDA
Nippon Telegraph and Telephone Corporation
Seishi TAKAMURA
Nippon Telegraph and Telephone Corporation
Hideaki KIMATA
Nippon Telegraph and Telephone Corporation
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
부
Shinobu KUDO, Shota ORIHASHI, Ryuichi TANIDA, Seishi TAKAMURA, Hideaki KIMATA, "GAN-Based Image Compression Using Mutual Information for Optimizing Subjective Image Similarity" in IEICE TRANSACTIONS on Information,
vol. E104-D, no. 3, pp. 450-460, March 2021, doi: 10.1587/transinf.2020EDP7080.
Abstract: Recently, image compression systems based on convolutional neural networks that use flexible nonlinear analysis and synthesis transformations have been developed to improve the restoration accuracy of decoded images. Although these methods that use objective metric such as peak signal-to-noise ratio and multi-scale structural similarity for optimization attain high objective results, such metric may not reflect human visual characteristics and thus degrade subjective image quality. A method using a framework called a generative adversarial network (GAN) has been reported as one of the methods aiming to improve the subjective image quality. It optimizes the distribution of restored images to be close to that of natural images; thus it suppresses visual artifacts such as blurring, ringing, and blocking. However, since methods of this type are optimized to focus on whether the restored image is subjectively natural or not, components that are not correlated with the original image are mixed into the restored image during the decoding process. Thus, even though the appearance looks natural, subjective similarity may be degraded. In this paper, we investigated why the conventional GAN-based compression techniques degrade subjective similarity, then tackled this problem by rethinking how to handle image generation in the GAN framework between image sources with different probability distributions. The paper describes a method to maximize mutual information between the coding features and the restored images. Experimental results show that the proposed mutual information amount is clearly correlated with subjective similarity and the method makes it possible to develop image compression systems with high subjective similarity.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2020EDP7080/_p
부
@ARTICLE{e104-d_3_450,
author={Shinobu KUDO, Shota ORIHASHI, Ryuichi TANIDA, Seishi TAKAMURA, Hideaki KIMATA, },
journal={IEICE TRANSACTIONS on Information},
title={GAN-Based Image Compression Using Mutual Information for Optimizing Subjective Image Similarity},
year={2021},
volume={E104-D},
number={3},
pages={450-460},
abstract={Recently, image compression systems based on convolutional neural networks that use flexible nonlinear analysis and synthesis transformations have been developed to improve the restoration accuracy of decoded images. Although these methods that use objective metric such as peak signal-to-noise ratio and multi-scale structural similarity for optimization attain high objective results, such metric may not reflect human visual characteristics and thus degrade subjective image quality. A method using a framework called a generative adversarial network (GAN) has been reported as one of the methods aiming to improve the subjective image quality. It optimizes the distribution of restored images to be close to that of natural images; thus it suppresses visual artifacts such as blurring, ringing, and blocking. However, since methods of this type are optimized to focus on whether the restored image is subjectively natural or not, components that are not correlated with the original image are mixed into the restored image during the decoding process. Thus, even though the appearance looks natural, subjective similarity may be degraded. In this paper, we investigated why the conventional GAN-based compression techniques degrade subjective similarity, then tackled this problem by rethinking how to handle image generation in the GAN framework between image sources with different probability distributions. The paper describes a method to maximize mutual information between the coding features and the restored images. Experimental results show that the proposed mutual information amount is clearly correlated with subjective similarity and the method makes it possible to develop image compression systems with high subjective similarity.},
keywords={},
doi={10.1587/transinf.2020EDP7080},
ISSN={1745-1361},
month={March},}
부
TY - JOUR
TI - GAN-Based Image Compression Using Mutual Information for Optimizing Subjective Image Similarity
T2 - IEICE TRANSACTIONS on Information
SP - 450
EP - 460
AU - Shinobu KUDO
AU - Shota ORIHASHI
AU - Ryuichi TANIDA
AU - Seishi TAKAMURA
AU - Hideaki KIMATA
PY - 2021
DO - 10.1587/transinf.2020EDP7080
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E104-D
IS - 3
JA - IEICE TRANSACTIONS on Information
Y1 - March 2021
AB - Recently, image compression systems based on convolutional neural networks that use flexible nonlinear analysis and synthesis transformations have been developed to improve the restoration accuracy of decoded images. Although these methods that use objective metric such as peak signal-to-noise ratio and multi-scale structural similarity for optimization attain high objective results, such metric may not reflect human visual characteristics and thus degrade subjective image quality. A method using a framework called a generative adversarial network (GAN) has been reported as one of the methods aiming to improve the subjective image quality. It optimizes the distribution of restored images to be close to that of natural images; thus it suppresses visual artifacts such as blurring, ringing, and blocking. However, since methods of this type are optimized to focus on whether the restored image is subjectively natural or not, components that are not correlated with the original image are mixed into the restored image during the decoding process. Thus, even though the appearance looks natural, subjective similarity may be degraded. In this paper, we investigated why the conventional GAN-based compression techniques degrade subjective similarity, then tackled this problem by rethinking how to handle image generation in the GAN framework between image sources with different probability distributions. The paper describes a method to maximize mutual information between the coding features and the restored images. Experimental results show that the proposed mutual information amount is clearly correlated with subjective similarity and the method makes it possible to develop image compression systems with high subjective similarity.
ER -