深度學習與圖像復原

第1 章 基於傳統機器學習的圖像復原方法 ............................................................. 1
1.1 圖像去噪 ···············································································1
1.1.1 圖像去噪任務簡介···························································1
1.1.2 基於傳統機器學習的圖像去噪方法 ·····································1
1.2 圖像超分辨率 ·········································································9
1.2.1 圖像超分辨率任務簡介 ····················································9
1.2.2 基於傳統機器學習的圖像超分辨率方法 ·······························9
1.3 圖像去水印 ·········································································.15
1.3.1 圖像去水印任務簡介 ····················································.15
1.3.2 基於傳統機器學習的圖像去水印方法 ·······························.15
1.4 本章小結 ············································································.19
參考文獻 ···················································································.20
第2 章 基於捲積神經網絡的圖像復原方法基礎 ................................................... 24
2.1 捲積層 ···············································································.24
2.1.1 捲積操作 ····································································.26
2.1.2 感受野 ·······································································.29
2.1.3 多通道捲積和多捲積核捲積 ···········································.30
2.1.4 空洞捲積 ····································································.31
2.2 激活層 ···············································································.33
2.2.1 Sigmoid 激活函數 ·························································.33
2.2.2 Softmax 激活函數 ·························································.35
2.2.3 ReLU 激活函數 ···························································.36
2.2.4 Leaky ReLU 激活函數 ···················································.38
2.3 基於捲積神經網絡的圖像去噪方法 ···········································.39
2.3.1 研究背景 ····································································.39
2.3.2 網絡結構 ····································································.40
2.3.3 實驗結果 ····································································.42
2.3.4 研究意義 ····································································.47
2.4 基於捲積神經網絡的圖像超分辨率方法 ·····································.48
2.4.1 研究背景 ····································································.48
2.4.2 網絡結構 ····································································.48
2.4.3 實驗結果 ····································································.51
2.4.4 研究意義 ····································································.55
2.5 基於捲積神經網絡的圖像去水印方法 ········································.55
2.5.1 研究背景 ····································································.55
2.5.2 網絡結構 ····································································.56
2.5.3 實驗結果 ····································································.58
2.5.4 研究意義 ····································································.61
2.6 本章小結 ············································································.62
參考文獻 ···················································································.62
第3 章 基於雙路徑捲積神經網絡的圖像去噪方法 ............................................... 69
3.1 引言 ··················································································.69
3.2 相關技術 ············································································.70
3.2.1 空洞捲積技術 ······························································.70
3.2.2 殘差學習技術 ······························································.71
3.3 面向圖像去噪的雙路徑捲積神經網絡 ········································.72
3.3.1 網絡結構 ····································································.72
3.3.2 損失函數 ····································································.74
3.3.3 重歸一化技術、空洞捲積技術和殘差學習技術的結合利用 ····.74
3.4 實驗結果與分析 ···································································.76
3.4.1 實驗設置 ····································································.77
3.4.2 關鍵技術的合理性和有效性驗證 ·····································.79
3.4.3 灰度與彩色高斯噪聲圖像去噪 ········································.83
3.4.4 真實噪聲圖像去噪························································.87
3.4.5 去噪網絡的復雜度及運行時間 ········································.89
3.5 本章小結 ············································································.89
參考文獻 ···················································································.90
第4 章 基於註意力引導去噪捲積神經網絡的圖像去噪方法 ............................... 93
4.1 引言 ··················································································.93
4.2 註意力方法介紹 ···································································.94
4.3 面向圖像去噪的註意力引導去噪捲積神經網絡 ···························.94
4.3.1 網絡結構 ····································································.95
4.3.2 損失函數 ····································································.96
4.3.3 稀疏機制和特徵增強機制 ··············································.96
4.3.4 註意力機制和重構機制 ·················································.98
4.4 實驗與分析 ·········································································.99
4.4.1 實驗設置 ····································································.99
4.4.2 稀疏機制的合理性和有效性驗證 ···································.100
4.4.3 特徵增強機制和註意力機制的合理性和有效性驗證 ···········.102
4.4.4 定量和定性分析 ·························································.103
4.5 本章小結 ···········································································.110
參考文獻 ··················································································.110
第5 章 基於級聯捲積神經網絡的圖像超分辨率方法 ......................................... 114
5.1 引言 ·················································································.114
5.2 相關技術 ···········································································.115
5.2.1 基於級聯結構的深度捲積神經網絡 ·································.115
5.2.2 基於模塊深度捲積神經網絡的圖像超分辨率 ·····················.116
5.3 面向圖像超分辨率的模塊深度捲積神經網絡 ······························.117
5.3.1 網絡結構 ···································································.118
5.3.3 低頻結構信息增強機制 ················································.119
5.3.4 信息提純塊 ·······························································.120
5.3.5 與主流網絡的相關性分析 ············································.121
5.4 實驗與分析 ·······································································.123
5.4.1 實驗設置 ··································································.123
5.4.2 特徵提取塊和增強塊的合理性和有效性驗證 ····················.124
5.4.3 構造塊和特徵細化塊的合理性和有效性驗證 ····················.126
5.4.4 定量和定性估計 ·························································.127
5.5 本章小結 ··········································································.135
參考文獻 ·················································································.136
第6 章 基於異構組捲積神經網絡的圖像超分辨率方法 ..................................... 142
6.1 引言 ················································································.142
6.2 相關技術 ··········································································.143
6.2.1 基於結構特徵增強的圖像超分辨率方法 ··························.143
6.2.2 基於通道增強的圖像超分辨率方法 ································.144
6.3 面向圖像超分辨率的異構組捲積神經網絡 ································.145
6.3.1 網絡結構 ··································································.145
6.3.2 損失函數 ··································································.147
6.3.3 異構組塊 ··································································.148
6.3.4 多水平增強機制 ·························································.149
6.3.5 並行上採樣機制 ·························································.150
6.4 實驗結果與分析 ·································································.155
6.4.1 數據集 ·····································································.155
6.4.2 實驗設置 ··································································.155
6.4.3 方法分析 ··································································.156
6.4.4 實驗結果 ··································································.157
6.5 本章小結 ··········································································.166
參考文獻 ·················································································.166
第7 章 基於自監督學習的圖像去水印方法 ......................................................... 173
7.1 引言 ················································································.173
7.2 自監督學習 ·······································································.174
7.2.1 捲積神經網絡 ····························································.175
7.2.2 生成對抗網絡 ····························································.176
7.2.3 註意力機制 ·······························································.176
7.2.4 混合模型 ··································································.176
7.3 面向圖像去水印的自監督學習方法 ·········································.177
7.3.1 基於自監督捲積神經網絡的結構 ···································.177
7.3.2 異構網絡 ··································································.178
7.3.3 感知網絡 ··································································.179
7.3.4 損失函數 ··································································.179
7.4 實驗結果與分析 ·································································.180
7.4.1 數據集 ·····································································.180
7.4.2 實驗設置 ··································································.180
7.4.3 方法分析 ··································································.181
7.4.4 實驗結果 ··································································.184
7.5 本章小結 ··········································································.189
參考文獻 ·················································································.189
第8 章 總結與展望 ................................................................................................ 195
8.1 總結 ················································································.195
8.2 展望 ················································································.197
致謝 ............................................................................................................................. 198