《電子技術(shù)應(yīng)用》
您所在的位置:首頁(yè) > 其他 > 設(shè)計(jì)應(yīng)用 > 基于生成對(duì)抗網(wǎng)絡(luò)的無(wú)監(jiān)督圖像超分辨率算法
基于生成對(duì)抗網(wǎng)絡(luò)的無(wú)監(jiān)督圖像超分辨率算法
信息技術(shù)與網(wǎng)絡(luò)安全 1期
趙志博,滕奇志,任 超,何小海,翟 森
(四川大學(xué) 電子信息學(xué)院,四川 成都610065)
摘要: 目前,大多數(shù)基于學(xué)習(xí)的圖像超分辨率研究通常采用預(yù)定的降質(zhì)類(lèi)型(比如雙三次下采樣)處理高分辨率圖像,來(lái)產(chǎn)生成對(duì)的訓(xùn)練集。然而,真實(shí)圖像往往存在未知的模糊和噪聲,導(dǎo)致這些算法無(wú)法有效應(yīng)用到真實(shí)場(chǎng)景中。為了實(shí)現(xiàn)真實(shí)圖像的超分辨率重建,提出了一種基于生成對(duì)抗網(wǎng)絡(luò)的無(wú)監(jiān)督圖像超分辨率算法,所提出的算法分為域轉(zhuǎn)換子網(wǎng)絡(luò)和重建子網(wǎng)絡(luò)兩個(gè)部分。同時(shí)設(shè)計(jì)了深度特征提取模塊,通過(guò)融合不同感受野所提取的圖像特征來(lái)提升網(wǎng)絡(luò)的性能。實(shí)驗(yàn)結(jié)果證明,相比于目前多數(shù)的圖像超分辨率算法,本文算法能夠?qū)崿F(xiàn)真實(shí)降質(zhì)圖像(存在噪聲、模糊等)的圖像超分辨率,在主觀效果和客觀指標(biāo)上均能獲得更好的性能。
中圖分類(lèi)號(hào): TP183;TP391
文獻(xiàn)標(biāo)識(shí)碼: A
DOI: 10.19358/j.issn.2096-5133.2022.01.009
引用格式: 趙志博,滕奇志,任超,等. 基于生成對(duì)抗網(wǎng)絡(luò)的無(wú)監(jiān)督圖像超分辨率算法[J].信息技術(shù)與網(wǎng)絡(luò)安全,2022,41(1):55-62.
Unsupervised image super-resolution algorithm based on Generative Adversarial Network
Zhao Zhibo,Teng Qizhi,Ren Chao,He Xiaohai,Zhai Sen
(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China)
Abstract: In most existing researches on learning-based image super-resolution, the pair of training datasets is generated by down-scaling high-resolution(HR) images through a predetermined operation(e.g.,bicubic down-sampling). However, these algorithms cannot be effectively applied to real scenes since the real-world image contains unknown noise and blur. To this end, we propose an unsupervised image super-resolution algorithm based on Generative Adversarial Network in this paper. Our method contains two parts: domain conversion sub-network and reconstruction sub-network. In addition, the deep feature extraction module is proposed to improve the performance of the network by merging the image features captured by different receptive fields. Extensive experiments illustrate that compared with most current image super-resolution algorithms, the proposed method can be applied to real-world image (containing noise, blur, etc.) super-resolution, and achieves the start-of-the-art(SOTA) performance on both subjective and objective evaluations.
Key words : real-world image super-resolution;domain conversion;Generative Adversarial Network;unsupervised training

0 引言

圖像是信息的重要載體,隨著數(shù)字圖像在醫(yī)學(xué)、監(jiān)控、遙感等領(lǐng)域的迅速發(fā)展,人們對(duì)圖像質(zhì)量的要求也越來(lái)越高。然而在實(shí)際的圖像獲取過(guò)程中,比如在視頻監(jiān)控領(lǐng)域,由于成像設(shè)備的限制,無(wú)法獲得滿足實(shí)際需求的更高空間分辨率的圖像,不利于后續(xù)對(duì)圖像信息的進(jìn)一步分析。同時(shí),在成像過(guò)程中由于受到成像條件等一系列因素影響,導(dǎo)致獲取的圖像存在一定程度的模糊和噪聲,顯著影響了圖像的質(zhì)量。圖像超分辨率重建技術(shù)可以在不需要改變現(xiàn)有成像設(shè)備等條件的前提下,根據(jù)低質(zhì)量(Low Quality,LQ)圖像重建出理想的高質(zhì)量(High Quality,HQ)圖像,在成本、實(shí)時(shí)性以及便利性等方面具有顯著的優(yōu)勢(shì),已經(jīng)成為了數(shù)字圖像處理技術(shù)的主要研究?jī)?nèi)容。一般來(lái)說(shuō),LQ圖像的退化模型可以描述為:

y=Px+n(1)

其中,y和x分別表示LQ圖像與對(duì)應(yīng)的HQ圖像,P表示圖像的退化矩陣,n代表圖像噪聲。因此,如果要重建出理想的HQ圖像,必須綜合考慮模糊和噪聲等影響圖像質(zhì)量的因素。



本文詳細(xì)內(nèi)容請(qǐng)下載:http://ihrv.cn/resource/share/2000003937。






作者信息:

趙志博,滕奇志,任  超,何小海,翟  森

(四川大學(xué) 電子信息學(xué)院,四川 成都610065)


此內(nèi)容為AET網(wǎng)站原創(chuàng),未經(jīng)授權(quán)禁止轉(zhuǎn)載。