從計(jì)算機(jī)視覺到圖像理解和多媒體分析,超像素得到學(xué)術(shù)界越來越多的關(guān)注。超像素生成是指根據(jù)局部像素的顏色、位置等特征對(duì)其進(jìn)行分組。簡(jiǎn)而言之,同一超像素內(nèi)的所有像素都具有相似或相同的特征。與像素相比,超像素包含更多的局部信息。此外,它們可以保持圖像中大部分物體的邊界。它為計(jì)算圖像特征和降低后續(xù)圖像處理任務(wù)的復(fù)雜度提供了一種簡(jiǎn)便的方法,從而引起了人們對(duì)計(jì)算機(jī)視覺的濃厚興趣,同時(shí)它還可以顯著地提高算法的處理效率。一個(gè)比較熱門的研究方向是超像素分割,近年來已經(jīng)提出了很多算法。這些算法使用不同的性能評(píng)估指標(biāo)和數(shù)據(jù)集進(jìn)行評(píng)估,從而導(dǎo)致算法之間的比較存在差異,這給研究人員比較前沿算法和評(píng)價(jià)其優(yōu)缺點(diǎn)提供了一個(gè)參考。這些算法大多在無噪聲的自然圖像上進(jìn)行計(jì)算,得到了很好的結(jié)果。然而,目前針對(duì)自然圖像中常見的噪聲情況,還沒有人對(duì)超像素分割算法的魯棒性進(jìn)行全面的研究。本文針對(duì)常見的噪聲類型,研究分析了超像素分割算法的魯棒性。本文的研究主要分為兩個(gè)重要階段。在研究的第一階段也是最重要的階段,對(duì)最近提出的11種超像素分割算法面對(duì)不同類型噪聲的魯棒性進(jìn)行了性能評(píng)估,并在此基礎(chǔ)上提出了相應(yīng)的算法。為此,選取不同程度的二維... 

【文章來源】：山東大學(xué)山東省 211工程院校 985工程院校 教育部直屬院校

【文章頁(yè)數(shù)】：128 頁(yè)

【學(xué)位級(jí)別】：博士

【文章目錄】：
Abstract in Chinese
Abstract in English
Chapter 1 Introduction
    1.1 Introduction to Superpixels
    1.2 Background and Motivation
    1.3 Research Overview
    1.4 Primary work and Contribution
    1.5 Outline of the Thesis
Chapter 2 Literature Review
    2.1 A review of commonly used Superpixel Segmentation algorithms
    2.2 Role of Superpixel Segmentation
    2.3 Categories of Superpixel segmentation algorithms
        2.3.1 Clustering based techniques
        2.3.2 Graph-based techniques
        2.3.3 Geometric flow based techniques
    2.4 Conclusion
Chapter 3 Superpixel Segmentation Algorithms
    3.1 Clustering based techniques
        3.1.1 Simple Linear Iterative Clustering (SLIC)
        3.1.2 VCells
        3.1.3 Manifold-SLIC (M-SLIC)
        3.1.4 Real-time superpixel segmentation by a DBSCAN clustering algorithm
        3.1.5 Superpixels Extracted via Energy-Driven Sampling (SEEDS)
    3.2 Graph-based technique
        3.2.1 Lazy Random Walks (LRW)
        3.2.2 Partially absorbing random walks (PARW)
    3.3 Geometric flow based techniques
        3.3.1 Bilateral geodesic algorithm (Bilateral-G)
        3.3.2 Flooding-based superpixel generation (FCCS)
        3.3.3 Structure Sensitive Superpixel via Geodesic distance (SSS-G)
        3.3.4 Turbopixel (TP)
    3.4 Conclusion
Chapter 4 Data set,Benchmarks and Types of Noise
    4.1 Berkeley Segmentation Data-set
    4.2 Evaluation Criteria
        4.2.1 Performance Evaluation Parameters
        4.2.2 Achievable Segmentation Accuracy (ASA)
        4.2.3 Under-segmentation Error (USE)
        4.2.4 Compactness
        4.2.5 Boundary Recall (BR)
    4.3 Types of noise
        4.3.1 2D Gaussian blur
        4.3.2 Additive white Gaussian noise (AWGN)
        4.3.3 Impulse noise
    4.4 Conclusion
Chapter 5 Robustness analysis of Superpixel Segmentation Algorithms
    5.1 Introduction
    5.2 Superpixel Segmentation Algorithms
    5.3 Quantitative Evaluation Measures
    5.4 Experiments and Results
        5.4.1 Evaluation Process
        5.4.2 Experimental Setup
        5.4.3 Experiments on Robustness to 2D-Gaussian Blur
        5.4.4 Experiments on Robustness to Additive White Gaussian Noise(AWGN)
        5.4.5 Experiments on Robustness to Impulse Noise
        5.4.6 Percentage Performance Degradation Analysis
        5.4.7 Rate of Performance Degradation Analysis
    5.5 Conclusion
Chapter 6 Experimental approach for evaluation of superpixels consistency
    6.1 Introduction
    6.2 Problem statement and algorithm Overview
    6.3 Algorithm Details
        6.3.1 Jaccard Similarity Coefficient (JSC)
        6.3.2 Data set and Noise
        6.3.3 Superpixel Segmentation algorithms
    6.4 Our Approach
        6.4.1 Computation of Superpixels
        6.4.2 Computation of similarity indices
        6.4.3 Determination of the coefficient threshold τ
        6.4.4 Algorithm 1
        6.4.5 Final grouping (Output)
    6.5 Experiments and Results
        6.5.1 Experimental Setup
        6.5.2 Experiment One (2D Gaussian Blur)
        6.5.3 Experiment Two (Impulse Noise)
        6.5.4 Experiment Three(2D Gaussian Blur + Impulse Noise)
    6.6 Conclusion
Chapter 7 Conclusion and Future Work
    7.1 Important Observations and Findings
    7.2 List of Recommendations
    7.3 Future Work
References
Acknowledgements
List of Published Papers
學(xué)位論文評(píng)閱及答辯情況表


【參考文獻(xiàn)】：
期刊論文
[1]Saliency detection based on superpixels clustering and stereo disparity[J]. GAO Shan-shan,CHI Jing,LI Li,ZOU Ji-biao,ZHANG Cai-ming.  Applied Mathematics:A Journal of Chinese Universities. 2016(01)
[2]Edge-Weighted Centroidal Voronoi Tessellations[J]. Jie Wang and Xiaoqiang Wang~* Department of Scientific Computing,Florida State University,Tallahassee, FL 532306-4120,USA..  Numerical Mathematics:Theory,Methods and Applications. 2010(02)



本文編號(hào)：3474342