指導教授：貝蘇章臺灣大學：電信工程學研究所張雯雯Chang, Wen-WenWen-WenChang2014-11-302018-07-052014-11-302018-07-052014http://ntur.lib.ntu.edu.tw//handle/246246/264251顯著性偵測(人類視覺注意力偵測)，指的是人類觀測者第一眼會注意到的區域，這些區域和其周圍區域通常有著顯著的差異。顯著性偵測對很多電腦視覺上應用上有幫助，近年來很多研究致力於提升顯著性偵測的準確性。經由觀察發現大部分現有的偵測方法，都難以消去複雜的背景區域並偵測出完整的顯著物體。因此我們運用將影像邊緣區域視作背景的假設及考慮各區域在空間上的關連性以提升準確度。 在本篇論文中，我們提出一個新的影像顯著性偵測方法，並將其延伸至影片顯著性偵測。此方法是基於假設影像邊緣為背景，和一個以超像素為節點單位的馬可夫隨機場模型。首先將影像分割成超像素，再取出每個超像素內的色彩、紋理、聚焦程度的特徵值。然後建立一個馬可夫隨機場模型來描述空間和時間上節點之間的關連性。最後再將以超像素為單位的顯著圖轉換為以像素為單位的顯著圖。實驗結果證實我們的方法可以準確的提取出影像和影片中的顯著性物體，並優於大部分現有的方法。Saliency, also known as visual attention, refers to the areas distinct from its surroundings that human observer would focus at a glance. Saliency detection benefits many computer vision tasks, and extensive efforts have been devoted to achieving better saliency detection performance. We observe that most of the previous works are hard to deal with the non-homogeneous color distribution within an object. Motivated by this observation, we consider the spatial structure between image regions to obtain better results. In this thesis, a proposed approach for image saliency detection and its extension for video saliency detection are introduced. The approach is based on background prior and superpixel-level Markov Random Field (MRF) model. First, we separate the image into middle-level superpixels and extract low-level features (color, texture energy, and defocus level) within each superpixel. Then, we build up a Markov-Random-Field (MRF) on the superpixels and adopt simplified propagation technique to optimize the superpixel saliency. Afterward, we refine this superpixel-level solution to pixel-level saliency map. Experimental results demonstrate that our proposed method is promising as compared to the state-of-the-art methods in two public available datasets.口試委員會審定書 誌謝 i 中文摘要 ii ABSTRACT iii CONTENTS iv LIST OF FIGURES vii Chapter 1 Introduction 1 1.1 What Is Saliency 1 1.2 Applications 2 1.3 Organization 3 Chapter 2 Overview of Previous Works on Image Saliency Detection 4 2.1 Biologically Based Approach 4 2.1.1 L. Itti’s Model 4 2.1.2 Simulation and Discussion 6 2.2 Frequency Domain Approaches 8 2.2.1 Spectral Residual (SR) 8 2.2.2 Phase Spectrum of Quaternion Fourier Transform (PQFT) 9 2.2.3 Hypercomplex Fourier Transform (HFT) 12 2.2.4 Simulation and Discussion 13 2.3 Context-Aware Saliency 15 2.3.1 Context-Aware Saliency 15 2.3.2 Simulation and Discussion 18 2.4 Global Contrast Approach 19 2.4.1 Histogram Based Contrast (HC) 19 2.4.2 Region Based Contrast (RC) 21 2.4.3 Simulation Results and Discussion 23 Chapter 3 Proposed Method for Image Saliency 24 3.1 Introduction 25 3.2 Superpixel Segmentation 27 3.3 Background Contrast 29 3.4 Probabilistic Model 32 3.4.1 Markov Random Field Model 32 3.4.2 Data Term Energy and Smoothness Term Energy 34 3.4.3 Optimization Using Belief Propagation 38 3.4.4 Optimization Using a Simpler Method 41 3.4.5 Methodology Evaluation 46 3.5 Refinement using Guided Filter 48 3.6 Performance Evaluation 51 3.6.1 Database 51 3.6.2 Performance Evaluation Methods 51 3.6.3 Quantitative Results 52 3.6.4 Qualitative Results 53 3.7 Applications on Image Retargeting 55 3.7.1 Image Retargeting Algorithm 55 3.7.2 Performance 56 Chapter 4 Overview of Previous Works on Video Saliency Detection 58 4.1 Phase Discrepancy 58 4.1.1 Algorithm 58 4.1.2 Simulation and Discussion 61 4.2 Spatial Temporal Spectral 62 4.2.1 Algorithm 62 4.2.2 Simulation and Discussion 65 4.3 Saliency-Based Video Object Extraction Using CRF Model 66 Chapter 5 Proposed Method for Video Saliency 70 5.1 Introduction 70 5.2 Video Saliency Detection Using Proposed Image Saliency Detection Algorithm 72 5.3 Superpixel Segmentation 74 5.4 Motion Feature: Optical Flow 75 5.5 Background Contrast 77 5.6 Probabilistic Model 82 5.6.1 Spatial Temporal Markov Random Field Model 82 5.6.2 Superpixel Tracking 84 5.6.3 Optimization 85 5.6.4 Methodology Evaluation 86 5.7 Results of Our Proposed Approach 87 Chapter 6 Conclusion and Future Work 90 6.1 Conclusion 90 6.2 Future Work 91 REFERENCE 926145698 bytesapplication/pdf論文公開時間：2016/07/22論文使用權限：同意無償授權顯著性偵測視覺注意力超像素馬可夫隨機場thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/264251/1/ntu-103-R01942035-1.pdf