臺灣博碩士論文加值系統

影像分類在現在是一個熱門的議題，而把數千數萬張的影像分類不是一件容易的工作；分類系統是對影像的每個類別做特徵影像分析，以此為基礎結合影像內容中的色彩特徵，並且應用於分析影像內容以協助影像分類的工作。
在本研究中，從資料庫中選出訓練樣本，以視覺字塊的方式提取影像特徵，並且將區塊依照特性分為巨觀字與微觀字兩種類型，將取出的視覺字訓練成巨觀字典與微觀視覺字典，由巨觀字典與微觀字典為基礎建立類別模型，每個模型獨特，不易混淆；而後以建立的類別模型為分類依據，已達成分類影像的目的。

This issue of image classification has received much attention recently; however, to classify huge amount of images into different categories is hard. For each category, classification system should first analyze features of image, which is described by visual words, and then the classification model is constructed. Finally, a probabilistic classifier is proposed to effectively category images.

The proposed algorithm is first collected training samples from the database, and extract image feature by visual patch. In addition, the patch divided into macro words and micro words according to patch content and then macro and micro visual dictionary is constructed. We then build classification models with representative and effective. Finally, the MAP based classifier is developed to classification image correctly. Simulation results show that the categorization scheme achieve surprising performance.

摘要 I
Abstract II
致謝 III
目錄 IV
圖目錄 VII
表目錄 IX
一、 緒論 1
1.1 問題描述 1
1.2 研究動機 2
1.3 論文架構 7
二、 文獻回顧與探討 8
2.1 視覺字 8
Keyblock區塊特徵 9
2.2 分類方法 10
2.2.1. 低階特徵(全域) 10
2.2.2. 低階特徵(子塊) 11
2.2.3. 語義物件(Semantic objects) 12
2.2.4. 線上Codebook重新加權(Online Codebook Reweighting) 13
三、 研究方法與步驟 15
3.1 建立視覺字典 19
3.1.1. 區塊特徵擷取 19
3.1.2. 區塊特徵訓練 20
3.1.3. 區分巨觀及微觀視覺字典 21
3.2 影像分類 24
3.2.1.類別模型訓練 26
3.2.2. 類別機率計算 31
四、 實驗數據及結果 36
4.1數據庫 36
4.2效能評估 38
4.3訓練模型評估 46
4.3.1模型差異 47
4.3.2模型差異造成的分類錯誤 49
4.3.3去除視覺字與否的差異測試 51
4.3.4視覺字使用評估 60
五、 結論與未來研究 66
參考文獻 67

[1]W. J. Xie, D. Xu, Y. J. Tang, and S. Y. Liu, “Constructing Class-Specific Codebooks based on Mutual Information Method for Natural Scene Categorization,” JOURNAL OF INFORMATION SCIENCE AND ENGINEERING XX, 1-xxx (xxxx).
[2]C.M.Kuo,M.H.Hung,C.S.Liu,Y.Chang and C.H.Hsieh,” Playfield Segmentation for Baseball Videos Using Adaptive GMMs,” International Journal of Innovative Computing, Information and Control, vol. 6, no. 6, pp. 2787–2801, 2010.
[3]M. H. Kolekar,” Bayesian belief network based broadcast sport video indexing,” Multimed Tools Appl, pp.27-54,2011.
[4]Y. Kang, A. Sugimoto,” Scale-Optimized Textons for Image Categorization and Segmentation,” IEEE International Symposium on Multimedia,2011.
[5]L.K Huang,” Visual Words With Scale-Invariant Features and Color Features for Image Description and Classification,” 2012.
[6]Alfanindya. A, Hashim. N and Eswaran. C,” Content Based Image Retrieval And Classification Using Speeded-Up Robust Features (SURF) and Grouped Bag-of-Visual-Words (GBoVW),” Technology, Informatics, Management, Engineering, and Environment (TIME-E),” 2013.
[7]Y.M.Chen, “Block-based Visual Words for Image Retrieval and Classification,”2012.
[8]A. Bosch, X. Mun ̃oz, and R. Martı´,” Which is the best way to organize/classify images by content?,” Image and Vision Computing, pp.778-791,2006.
[9]Xin Zhao, Weiqiang Ren, Kaiqi Huang, and Tieniu Tan, “Online Codebook Reweighting Using Pairwise Constraints for Image Classification,” 2011
[10]Y.S.Sie, “Image Categorization Based on Bag of Visual Words,”2014.