應(yīng)用于視覺(jué)測(cè)量的圖像超分辨率重建算法
信息技術(shù)與網(wǎng)絡(luò)安全 5期
王亞金,吳麗君,陳志聰,鄭 巧,程樹(shù)英,林培杰
(福州大學(xué) 物理與信息工程學(xué)院,福建 福州350108)
摘要: 惡劣環(huán)境下的低質(zhì)圖像會(huì)嚴(yán)重影響基于視覺(jué)的位移測(cè)量效果。圖像超分辨率重建有望能改善圖像質(zhì)量、突出目標(biāo)特征以提高測(cè)量精度和可靠性,進(jìn)而應(yīng)用于視覺(jué)測(cè)量場(chǎng)景。故提出了一種關(guān)注細(xì)節(jié)特征的圖像超分辨率重建算法,該算法設(shè)計(jì)了一個(gè)角點(diǎn)增強(qiáng)支路,并通過(guò)角點(diǎn)損失函數(shù)進(jìn)行約束實(shí)現(xiàn)對(duì)角點(diǎn)信息的增強(qiáng),此外增加邊緣損失函數(shù)提升邊緣的重建效果。實(shí)驗(yàn)結(jié)果表明,該算法在客觀(guān)評(píng)價(jià)指標(biāo)上表現(xiàn)優(yōu)異,視覺(jué)效果上取得了更加清晰的紋理細(xì)節(jié),設(shè)計(jì)的驗(yàn)證實(shí)驗(yàn)證明,該算法重建的邊緣與角點(diǎn)更加準(zhǔn)確,對(duì)目標(biāo)定位有一定幫助,適用于視覺(jué)測(cè)量應(yīng)用場(chǎng)景。
中圖分類(lèi)號(hào): TP391
文獻(xiàn)標(biāo)識(shí)碼: A
DOI: 10.19358/j.issn.2096-5133.2022.05.010
引用格式: 王亞金,吳麗君,陳志聰,等. 應(yīng)用于視覺(jué)測(cè)量的圖像超分辨率重建算法[J].信息技術(shù)與網(wǎng)絡(luò)安全,2022,41(5):66-71.
文獻(xiàn)標(biāo)識(shí)碼: A
DOI: 10.19358/j.issn.2096-5133.2022.05.010
引用格式: 王亞金,吳麗君,陳志聰,等. 應(yīng)用于視覺(jué)測(cè)量的圖像超分辨率重建算法[J].信息技術(shù)與網(wǎng)絡(luò)安全,2022,41(5):66-71.
Image super-resolution reconstruction algorithm for vision measurement
Wang Yajin,Wu Lijun,Chen Zhicong,Zheng Qiao,Cheng Shuying,Lin Peijie
(College of Physics and Information Engineering,F(xiàn)uzhou University,F(xiàn)uzhou 350108,China)
Abstract: The low-quality images in harsh enviroments seriously affect the effect of vision-based displacement measurement.Image super-resolution reconstruction is expected to improve image quality and highlight target features to improve measurement accuracy and reliability, and then applied to visual measurement scenarios. This study proposes an image super-resolution reconstruction algorithm that pays attention to detailed features. The algorithm designs a corner enhancement branch, and is constrained by the corner loss function to enhance the corner information, in addition to adding edge loss function. The loss function improves the reconstruction effect of the edge. The experimental results show that the algorithm performs well in objective evaluation indexes and achieves clearer texture details in visual effects. The designed verification experiment proves that the edges and corners reconstructed by the algorithm are more accurate, which is helpful for target positioning, and is suitable for visual measurement application scenarios.
Key words : deep learning;image processing;super-resolution;detailed features;visual measurement;displacement measurement;corner extration
0 引言
建筑結(jié)構(gòu)的位移監(jiān)測(cè)對(duì)于建筑的安全性保證是十分重要的,比如斜拉橋和懸索橋這類(lèi)采用索纜作為主體的建筑結(jié)構(gòu),在極端天氣下發(fā)生的振動(dòng)對(duì)整體建筑的安全性影響不可忽視。近年來(lái),基于視覺(jué)的非接觸式測(cè)量方法因?yàn)槠鋵?shí)用性被廣泛應(yīng)用于位移測(cè)量,但目前的方法仍存在一些局限性。首先,當(dāng)拍攝距離較遠(yuǎn),需要提高測(cè)量精度時(shí),通常只能縮小視場(chǎng)范圍,測(cè)量精度與視場(chǎng)大小二者是相互矛盾的,也可以考慮采用多個(gè)設(shè)備同步測(cè)量,但這會(huì)大幅度提高成本。其次,橋梁索纜圖像的采集是在戶(hù)外條件下,因此采集到的圖像會(huì)受到各種噪聲干擾,這種低質(zhì)圖像會(huì)給后續(xù)的位移測(cè)量工作帶來(lái)困難。圖像超分辨率重建是一種有效提升圖像質(zhì)量、重建目標(biāo)細(xì)節(jié)特征的圖像處理方法,為解決上述提出的位移測(cè)量問(wèn)題提供了一個(gè)新的思路。
本文詳細(xì)內(nèi)容請(qǐng)下載:http://theprogrammingfactory.com/resource/share/2000004249
作者信息:
王亞金,吳麗君,陳志聰,鄭 巧,程樹(shù)英,林培杰
(福州大學(xué) 物理與信息工程學(xué)院,福建 福州350108)
此內(nèi)容為AET網(wǎng)站原創(chuàng),未經(jīng)授權(quán)禁止轉(zhuǎn)載。