由于噪聲和硬件的限制,低端圖像設(shè)備采集到的圖像和視頻并不理想。因此,許多文獻(xiàn)中都對這個問題提出了解決方法。超分辨率技術(shù)就是其中一種將圖像或視頻由低質(zhì)量重構(gòu)成高質(zhì)量的一種內(nèi)在的適定性問題。大多數(shù)已有的超分辨率重構(gòu)算法都不能完全保留一些重要的圖像特征,比如物體的邊緣和輪廓等,然而事實上人眼會對這些邊緣很敏感,并且它們也在目標(biāo)檢測等計算機(jī)視覺應(yīng)用領(lǐng)域會起到很大的作用。為解決過去方法中的問題,本文中我們提出了幾種基于非線性擴(kuò)散泛函的超分辨率算法。新方法能根據(jù)圖像特征自動調(diào)整正則化水平。具體來說,正則化在平坦區(qū)域較強(qiáng)以消除噪聲,在邊緣區(qū)域較弱以保護(hù)重要的圖像信息。這種基于圖像特征的方法使得我們的模型重建后的圖像信息更詳細(xì)。首先,我們的超分辨率算法基于Perona-Malik光滑泛函,其中的擴(kuò)散性部分含有以空間為變量的指數(shù)項,它隨標(biāo)準(zhǔn)化變化而變化。第二,我們引入了一種改進(jìn)的Charbonnier模型來描述超分辨率的適定性問題。這種方法能適應(yīng)諸如線性等方向擴(kuò)散,全偏差以及Charbonnier等不同的正則化模型,并且具有靈活性,并且能產(chǎn)生可觀的超分辨結(jié)果。第三,為了能同時提高圖像的空間分辨率和重構(gòu)頻率成分,我們引入Papoulis-Gerchberg算法。最后,對于超分辨問題我們得到一個新的正則化勢函數(shù)。為保證勢函數(shù)的凸性、光滑性和單調(diào)性,我們在參數(shù)中加了適當(dāng)?shù)募s束條件。這種勢函數(shù)可以使我們的超分辨模型達(dá)到更高的分辨率,這在以往的模型中是達(dá)不到的。新的重構(gòu)算法有很廣泛的應(yīng)用。例如,可以應(yīng)用于改進(jìn)醫(yī)學(xué)上血涂片的圖像質(zhì)量,準(zhǔn)確檢測并診斷瘧疾等疾病。本文中,可以將任意一種超分辨率算法嵌入到低端圖像采集設(shè)備(采集的圖像是低分辨率圖像)中,來增強(qiáng)輸入圖像的質(zhì)量,這樣既避免了昂貴的顯微鏡設(shè)備,同時保證了高準(zhǔn)確性。而傳統(tǒng)的自動檢測診斷方法需要昂貴的硬件,許多人都無法支付。實驗結(jié)果顯示,本文中的模型較最先進(jìn)的其它經(jīng)典方法更高級。通過多種圖像、視頻的仿真,本文方法的視覺效果和性能指標(biāo)(噪聲信號峰值比、邊緣和結(jié)構(gòu)相似性)更理想。
【學(xué)位單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級別】：博士
【學(xué)位年份】：2015
【中圖分類】：TP391.41
【文章目錄】：
摘要
Abstract
Chapter 1 Introduction
    1.1 Synopsis
    1.2 Background of the Study
        1.2.1 What is resolution?
        1.2.2 Super-resolution imaging
    1.3 Related works and their limitations
    1.4 Objectives of the Research
    1.5 Significance of the Study
    1.6 Thesis outline and contributions
Chapter 2 Multiframe super-resolution image degradation model
    2.1 Introduction
    2.2 Image degradation model
    2.3 Regularization of the multiframe super-resolution problem
        2.3.1 Basics of inverse problems
        2.3.2 Regularization
    2.4 Comparisons of the classical regularizing functionals
    2.5 Summary
Chapter 3 Super-resolution methods based on the variable exponent nonlin-ear diffusion models
    3.1 Introduction
    3.2 Motion estimation
    3.3 Proposed methods
        3.3.1 Super-resolution method based on the adaptive Perona-Malik diffu-sion model
        3.3.2 Super-resolution method based on the adaptive Charbonnier diffusionmodel
        3.3.3 Super-resolution method based on the non-standard anisotropic diffu-sion model
        3.3.4 Super-resolution method based on the adaptive Perona-Malik modeland Papoulis-Gerchberg algorithm
    3.4 Experiments
        3.4.1 Preliminaries
        3.4.2 Experiment 1: Edge detection
        3.4.3 Experiment 2: Image denoising
        3.4.4 Experiment 3: Super-resolution image reconstruction
    3.5 Results and discussions
        3.5.1 Experiment 1: Edge detection
        3.5.2 Experiment 2: Image denoising
        3.5.3 Experiment 3: Super-resolution image reconstruction
    3.6 Summary
Chapter 4 A noise suppressing and edge-preserving multiframe super-resolutionmethod
    4.1 Introduction
    4.2 Motion estimation
    4.3 Proposed smoothing energy functional
        4.3.1 Derivations and important properties
        4.3.2 Multiframe super-resolution process
        4.3.3 Invariance and the regularizing parameter
    4.4 Numerical implementation details
        4.4.1 Explicit scheme
        4.4.2 Additive Operator Splitting （AOS） scheme
    4.5 Experiments
        4.5.1 Preliminaries
        4.5.2 Experiment 1: Edge detection
        4.5.3 Experiment 2: Image denoising
        4.5.4 Experiment 3: Super-resolution image reconstruction
    4.6 Results and discussions
        4.6.1 Experiment 1: Edge detection
        4.6.2 Experiment 2: Image denoising
        4.6.3 Experiment 3: Super-resolution image reconstruction
    4.7 Summary
Chapter 5 Practical applications of the super-resolution methods
    5.1 Introduction
    5.2 Practical applications of the super-resolution methods
        5.2.1 Fusion of images
        5.2.2 Improving the spatial resolution of mammograms in X-Ray imaging
        5.2.3 Improving the quality of hyperspectral images
        5.2.4 Resolution enhancement of scenes on the web
        5.2.5 Zooming of regions of interest （ROI） in the scene
        5.2.6 Lowering the transmission costs of videos from television broadcast-ing stations
        5.2.7 Improving the quality of consumer images and videos
    5.3 Summary
結(jié)論
Conclusion
References
List of Publications
Acknowledgement
Resume


本文編號：2857947