(3.230.143.40) 您好!臺灣時間:2021/04/21 07:07
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果

詳目顯示:::

我願授權國圖
: 
twitterline
研究生:陳永明
研究生(外文):Yung-Ming Chen
論文名稱:應用類神經網路於二維影像之深度圖估測
論文名稱(外文):Depth Map Estimation for 2D Images using Artificial Neural Network
指導教授:沈岱範
學位類別:碩士
校院名稱:國立雲林科技大學
系所名稱:電機工程系碩士班
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2013
畢業學年度:101
語文別:中文
論文頁數:68
中文關鍵詞:深度圖類神經倒傳遞網路深度影像繪圖法預測訓練
外文關鍵詞:PredictionDIBRBPNNTrainingDepth Map
相關次數:
  • 被引用被引用:1
  • 點閱點閱:401
  • 評分評分:系統版面圖檔系統版面圖檔系統版面圖檔系統版面圖檔系統版面圖檔
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
較之二維影像,三維影像更能帶給人們有身歷其境之感。在2D至3D轉換上,深度圖(Depth Map)的產生扮演著關鍵的角色。現在的相關研究裡,深度圖估測大多數是透過啟發式(heuristic)演算法來產生。利用影像分類及深度線索來做為影像特徵之分析,上述這些方法需考慮較多的因素來產生深度圖。也有少數研究利用訓練的方式來產生深度圖。因此本論文嘗試提出一個應用類神經網路於二維影像之深度圖估測方法。
近年來類神經網路(Neural Network)技術已經相當成熟,本實驗室曾成功應用類神經倒傳遞網路(Back-propagation Network,BPN)做蛋白質二級結構預測,其準確率超過80%,成果優異。因此本論文受其啟發亦使用類神經網路做深度圖估測。在訓練過程,由單重及多重紋路影像切方塊像素作為輸入向量,而其已知之深度值做為相對應之輸出,來訓練類神經網路,其訓練結果是一組神經元的權重值(Weight)。在預測過程,載入已訓練好之神經元權重值,依輸入之二維紋路影像取出輸入向量,而得到相對應的深度值。
實驗結果顯示,當增加輸入向量重數(multi-resolution)、神經元數量及隱藏層層數時,其預測精準度提升(MAE值可達到1.0)。最後經由深度影像繪圖法(Depth-image-based rendering,DIBR)技術合成左右眼影像,透過3D顯示器即可呈現出立體視覺的效果。
事後文獻搜尋,目前只有史丹佛大學採用(Markov Random Field,MRF)訓練方式做range sensor深度圖估測。
Compared with two-dimensional images, three-dimensional images to bring people more immersive feeling. In the 2D to 3D conversion, the depth map generation plays a key role. Now the relevant study, most of the depth map estimation through heuristic algorithm to generate. Classification and use of image depth cues do for image feature analysis, these methods need to consider many factors to generate the depth map. There are few studies use training methods to generate the depth map. Therefore, this paper attempts to propose an application of neural networks in the two-dimensional image of the depth map estimation method.
In recent years, neural network technology is quite mature, our laboratory has successfully applied back-propagation neural network(BPNN) to make protein secondary structure prediction, the accuracy rate of over 80%, the result is excellent. Therefore, this paper Inspired also uses neural networks to estimate the depth map. In the training process, and multiple texture from a single re-cut block image pixels as input vector, and its value is known as the depth corresponding to the output, to train the neural network, the training result is a set of neurons weights. In the prediction process, load a good training weights of neurons, according to the input texture of the two-dimensional input vector video out, and get the corresponding depth values.
Experimental results show that when increasing the number of input vector multi-resolution, the number of neurons and the number of hidden layers of prediction accuracy upgrade (MAE value up to 1.0). Finally through the depth of the image drawing method (DIBR) synthesized left and right eye images through the 3D display to render a three-dimensional visual effect. Later literature search, only Stanford University used (MRF) training methods do range sensor depth map estimation.
摘要 i
Abstract ii
誌謝 iii
目錄 iv
表目錄 vi
圖目錄 vii
第一章 緒論 1
1.1研究背景 1
1.2研究動機與目標 1
1.3研究方法 2
1.4論文特色及主要貢獻 2
1.5各章提要 2
第二章 文獻回顧 3
2.1立體影像/視訊產生過程 3
2.2左右眼影像深度估測 4
2.3單張影像深度估測 6
2.3.1第一類(影像分類、深度線索) 6
2.3.1.1影像分類 6
2.3.1.2深度線索 7
2.3.2第二類(利用訓練方法) 8
2.4深度影像繪圖法 8
2.5類神經網路(Neural Network) 9
2.5.1類神經網路由來 9
2.5.2類神經網路簡介 10
2.5.3類神經網路應用 12
2.5.4類神經網路優缺點 13
2.6第二章節結論 13
第三章 使用類神經網路做深度圖預測 15
3.1 NN深度圖訓練與預測之系統架構 16
3.2定義輸入向量及目標值 17
3.3創建網路 20
3.3.1本論文倒傳遞網路的架構與參數設定 20
3.3.2創建前饋網路 23
3.4訓練網路 25
3.5定義新的輸入向量 25
3.6預測網路 26
3.7本章節結論 29
第四章 多重解析度深度預測 30
4.1雙重解析度預測 32
4.2三重解析度預測 35
4.3四重解析度預測 37
4.4實驗結果 39
4.4.1實驗一:多重解析度預測 39
4.4.2實驗二:增加神經元數量 42
4.4.3實驗三:多張影像深度圖預測 45
4.4.3.1 zig-zag scan 45
4.4.3.2頭尾相接 45
4.4.3.3多張影像深度圖預測 47
4.4.3.4修正深度圖 51
4.5深度圖品質評估 52
4.5.1主客觀品質評估方式 52
4.5.2深度影像繪圖法(DIBR) 53
4.5.3立體影像 54
4.6本章節結論 55
第五章 結論與未來可改進方向 56
參考文獻 57
[1]Jiang. Hao, Guo. Shuxu, Meng. Siming, and Luo. Xiaonan, "A Novel Depth Map Generation Method Based on K-Means Clustering and Depth Pattern Recognition," in Internet of Things (iThings/CPSCom), 2011 International Conference on and 4th International Conference on Cyber, Physical and Social Computing, 2011, pp. 638-643.
[2]Kuo. Tien-Ying, Lo. Yi-Chung, and Lin. Chia-Chin, "2D-to-3D conversion for single-view image based on camera projection model and dark channel model," in Acoustics, Speech and Signal Processing (ICASSP), 2012 IEEE International Conference on, 2012, pp. 1433-1436.
[3]Yang. Na-Eun, Lee. Ji Won, and Park. Rae-Hong, "Depth map generation from a single image using local depth hypothesis," in Consumer Electronics (ICCE), 2012 IEEE International Conference on, 2012, pp. 311-312.
[4]A. Klaus, M. Sormann, and K. Karner, "Segment-Based Stereo Matching Using Belief Propagation and a Self-Adapting Dissimilarity Measure," Pattern Recognition, 2006. ICPR 2006. 18th International Conference on, 2006, pp. 15-18.
[5]Fan. Yu-Cheng, Chen. Po-Wei, and Chiu. Yi-Chih, "A 2D-to-3D Image Conversion System Using Block Slope Pattern Based Vanishing Point Detection Technology," in Computer, Consumer and Control (IS3C), 2012 International Symposium on, 2012, pp. 321-324.
[6]Tsai. Yi-Min, Chang. Yu-Lin, and Chen. Liang-Gee, "Block-based Vanishing Line and Vanishing Point Detection for 3D Scene Reconstruction," in Intelligent Signal Processing and Communications, 2006. ISPACS ''06. International Symposium on, 2006, pp. 586-589.
[7]Han. Kyuseo and Hong. Kihyun, "Geometric and texture cue based depth-map estimation for 2D to 3D image conversion," in Consumer Electronics (ICCE), 2011 IEEE International Conference on, 2011, pp. 651-652.
[8]Chang. Yu-Lin, Fang. Chih-Ying, Ding. Li-Fu, Chen. Shao-Yi, and Chen. Liang-Gee, "Depth Map Generation for 2D-to-3D Conversion by Short-Term Motion Assisted Color Segmentation," in Multimedia and Expo, 2007 IEEE International Conference on, 2007, pp. 1958-1961.
[9]Cheng. Chao-Chung, Li. Chung-Te, and Chen. Liang-Gee, "A 2D-to-3D conversion system using edge information," in Consumer Electronics (ICCE), 2010 Digest of Technical Papers International Conference on, 2010, pp. 377-378.
[10]Po. Lai-Man, Xu. Xuyuan, Zhu. Yuesheng, Zhang. Shihang, Cheung. Kwok-Wai, and Ting. Chi-Wang, "Automatic 2D-to-3D video conversion technique based on depth-from-motion and color segmentation," in Signal Processing (ICSP), 2010 IEEE 10th International Conference on, 2010, pp. 1000-1003.
[11]Liu. Chao and L. Christopher, "Depth map estimation from motion for 2D to 3D conversion," in Electro/Information Technology (EIT), 2012 IEEE International Conference on, 2012, pp. 1-4.
[12]Cheng. Chao-Chung, Li. Chung-Te, and Chen. Liang-Gee, "A novel 2Dd-to-3D conversion system using edge information," Consumer Electronics, IEEE Transactions on, vol. 56, pp. 1739-1745, 2010.
[13]Ashutosh Saxena, H. Chung-Sung, and Y. Ng-Andrew, "3-D Depth Reconstruction from a Single Still Image," International Journal of Computer Vision (IJCV), Aug 2007.
[14]J. Konrad, Wang. Meng, and P. Ishwar, "2D-to-3D image conversion by learning depth from examples," in Computer Vision and Pattern Recognition Workshops (CVPRW), 2012 IEEE Computer Society Conference on, 2012, pp. 16-22.
[15] 賴昀暘,單視角室外影像產生3D影像之技術,國立台北科技大學電機工程
系碩士論文,中華民國九十九年七月
[16] 洪銘聰,多重解析度與應用視覺特性之影像分割演算法,國立雲林科技大學
電機工程系碩士論文,中華民國九十一年七月
[17] 鍾興弦,使用區段局限性及其統計特性於蛋白質二級結構預測,國立雲林科
技大學電機工程系碩士論文,中華民國九十三年七月
[18] 黃湘芸,可處理字幕視訊之即時2D至3D視訊轉換技術,國立中正大學電
機工程所碩士論文,中華民國九十九年七月
[19] http://www.dofpro.com/cgigallery.htm
[20] "Make3D", Available:http://make3d.cs.cornell.edu/code.html
[21] 類神經網路-MATLAB的應用(第三版),羅華強編著。
[22] 影像處理與電腦視覺(第5版),鍾國亮編著。
[23]Zhang. Liang and Wa. James. Tam, "Stereoscopic image generation based on depth images for 3D TV," Broadcasting, IEEE Transactions on, vol. 51, pp. 191-199, 2005.
[24] 類神經網路概述及實例,輔仁大學統計學系,謝邦昌教授著。
[25] http://www.nvidia.com.tw/
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
系統版面圖檔 系統版面圖檔