光場(chǎng)相機(jī)例如Lytro公司生產(chǎn)的光場(chǎng)相機(jī)以先拍照后聚焦而聞名。光場(chǎng)相機(jī)可以在一次拍照的過(guò)程中同時(shí)獲得空間信息和角度信息,它可以記錄光傳播的方向。由于光場(chǎng)相機(jī)可以拍攝豐富的信息,使得它在重聚焦、深度估計(jì)、圖像顯著性檢測(cè)、材料識(shí)別和顯微鏡下恢復(fù)透明物體方面展現(xiàn)突出的優(yōu)勢(shì)。雖然用光場(chǎng)數(shù)據(jù)進(jìn)行的深度估計(jì)比基于雙目立體視覺系統(tǒng)的深度估計(jì)更簡(jiǎn)單,更有效,但光場(chǎng)的深度估計(jì)仍然存在一些挑戰(zhàn)性的問(wèn)題。它的準(zhǔn)確度仍然受到遮擋和噪聲的影響。而且,當(dāng)我們使用卷積神經(jīng)網(wǎng)絡(luò)進(jìn)行對(duì)光場(chǎng)數(shù)據(jù)進(jìn)行深度估計(jì)的時(shí)候,光場(chǎng)數(shù)據(jù)的數(shù)目不足以達(dá)到用卷積神經(jīng)網(wǎng)絡(luò)來(lái)模擬光場(chǎng)數(shù)據(jù)和相對(duì)應(yīng)的深度關(guān)系時(shí)所需要的數(shù)目。光場(chǎng)數(shù)據(jù)還有一個(gè)自身的缺點(diǎn)就是光場(chǎng)的傳感器的數(shù)量是一定的,所以光場(chǎng)的分辨率是受傳感器及微透鏡陣列所限制的。和傳統(tǒng)的相機(jī)所拍攝的圖像的分辨率相比光場(chǎng)某一視角的圖片的分辨率要低得多,這使得光場(chǎng)的適用范圍受到限制。這篇論文的主要目的就是光場(chǎng)的深度估計(jì)和光場(chǎng)的超分辨。基于這個(gè)目標(biāo),首先我們采用支持向量機(jī)和以分割為引導(dǎo)的雙邊濾波來(lái)解決深度估計(jì)中的遮擋問(wèn)題,其次我們用濾波的結(jié)果來(lái)產(chǎn)生網(wǎng)絡(luò)訓(xùn)練所需的數(shù)據(jù)的深度的真實(shí)值,然后采用網(wǎng)絡(luò)來(lái)估計(jì)光場(chǎng)... 

【文章來(lái)源】：西安電子科技大學(xué)陜西省 211工程院校 教育部直屬院校

【文章頁(yè)數(shù)】：73 頁(yè)

【學(xué)位級(jí)別】：碩士

【文章目錄】：
ABSTRACT
摘要
Abbreviations Table
Chapter 1 Introduction
    1.1 Light Field Camera
        1.1.1 Light Field
        1.1.2 Light Field Camera
    1.2 Light Field Data
    1.3 Overview of Light Field Depth Estimation
    1.4 Overview of Light Field Super-resolution
    1.5 Contribution and Overview
        1.5.1 Contribution
        1.5.2 Overview
Chapter 2 Background
    2.1 Computational Photography
    2.2 Light Field Refocus
    2.3 Light Field Zoom
    2.4 Light Field Depth Estimation
        2.4.1 Epipolar Plane Image
        2.4.2 Depth From Defocus
        2.4.3 Depth From Correspondence
    2.5 Convolution Neural Networks For Super-Resolution
    2.6 Bilateral Filter and Joint bialteral filter
Chapter 3 Occlusion Robust Light Field Depth Estimation Using Segmentation Guid-ed Bilateral Filtering
    3.1 Introduction
    3.2 Proposed Method
        3.2.1 Refocus
        3.2.2 SVM Classification
        3.2.3 Occlusion Robust Depth Estimation
        3.2.4 Segmentation Guided Bilateral Filtering
    3.3 Experimental Results
    3.4 Conclusions
Chapter 4 Light Field Depth Estimation Based on Convolutional Neural Network
    4.1 Introduction
    4.2 Proposed Method
        4.2.1 Training Data Regeneration
        4.2.2 CNN Architecture
        4.2.3 Data Augmentation
        4.2.4 Depth Prediction
        4.2.5 Denoising in Depth
    4.3 Experimental Results
        4.3.1 Quantitative Evaluatio
        4.3.2 Qualitative Evaluation
        4.3.3 Limitation and future work
    4.4 Conclusion
Chapter 5 Light Field Image Super-resolution Based on Disparity Compensated Pre-diction
    5.1 Introduction
    5.2 Proposed Method
        5.2.1 Disparity Estimation
        5.2.2 View Synthesis
        5.2.3 Light Field Super-resolution
        5.2.4 Loss Function
    5.3 Experimental result
    5.4 Conclusion
Chapter 6 Summary and Future Work
    6.1 Summary
    6.2 Future Work
Reference
Acknowledgement
Biograph



本文編號(hào)：3174202