臺灣博碩士論文加值系統

傳統的圖片檢索方式主要是使用人工標記的後設資料，進行關鍵字查詢；但是關鍵字查詢會有詞意模糊的問題常會使得查詢結果過於發散；又因關鍵字本身描述力的限制，無法精確表述形狀、紋理等抽象概念，以至不能找出使用者想要的結果。本論文的研究是藉著使用者手繪的草稿進行查詢，以找出形狀相似的草稿或圖片。我們提出了將草稿轉換成環形特徵序列，並使用動態時間校準進行特徵序列比對的方法。我們並提出了偏移容忍值和懲罰值來限制匹配路徑的彎曲度，對動態時間校準進行最佳化；在無正規化和正規化特徵進行檢索可分別達到0.8288和0.9121的平均精確率。進一步，我們對角度取樣率、尺度變化、和旋轉角度變化分別做實驗，以了解並改進容錯能力。實驗結果顯示：角度取樣率約360 的時候可以達到不錯的效能，而適度減小取樣率也不會造成檢索效能明顯變差；對特徵進行正規化則可以改善因草稿尺度差異而造成的效能下降；旋轉的補償方法也有效提升檢索對於草稿旋轉的容錯能力。最後，我們將草稿搜索應用於以手繪草稿的方式來查詢圖片，發現本論文的方法可以找出形狀相似的物體。這樣的技術未來可用於個人手繪本搜尋、圖片瀏覽、或是圖片的自然檢索介面。

Conventional image retrieval uses the query of keywords, which are often extracted from meta data of the images and used for indexing. However, the retrieval results are usually unsatisfactory because keywords might be ambiguous or vague. In addition, such concepts as shape, color or texture can hardly by expressed by keywords precisely, so the retrieval system cannot fulfill users’ purposes once in a while. Sketch retrieval, which belongs to content-based image retrieval, is therefore a promising solution. The goal of this paper is to retrieve similar sketches or images through hand-drawn sketches. In our approach, a sketch is represented as a sequence of point sets, called radial feature, and two sketches can be compared through dynamic time warping algorithm in which the degree of warping can be constrained by angular shift and penalty. In the basic experiment for 40 test sketches and 100 target sketches, 0.8288 and 0.9121 of mean average precision can be achieved for features without and with normalization, respectively. Other experiments further indicate that optimal retrieval performance can be obtained with angular sampling rate of 360, which could decrease without degrading the performance seriously. In addition, this approach was verified to be error tolerant for scaling and rotation. It was finally applied to the retrieval of 1,000 real images that were preprocessed and converted into radial features in advance. Experimental results show that this approach is effect and robust for finding out objects with similar shapes, and is potentially applicable to such applications as image browsing or sketch books.

第一章、序論 1
1.1研究動機 1
1.2論文主要成果 3
1.3論文組織與架構 4
第二章、文獻探討 5
2.1基於內容的圖像檢索 5
2.2相關研究 6
2.3 動態時間校準 10
2.4本章摘要 11
第三章、草稿搜索方法 12
3.1草稿資料表示法 12
3.2 環形特徵擷取 13
3.3 彈性比對方法 15
3.4效能度量 18
3.5 本章摘要 19
第四章、手繪草稿搜尋草稿 20
4.1實驗資料 20
4.2動態時間校準最佳化 21
4.3 特徵正規化對效能的影響 23
4.4不同比對方式的實驗 24
4.5 角度取樣率對效能的影響 25
4.6 縮放容錯性測試 26
4.7 旋轉容錯性測試 27
4.8本章摘要 29
第五章、手繪草稿搜尋圖片 30
5.1 圖片處理 30
5.2 實驗與分析 33
第六章、結論和未來研究方向 36
6.1結論 36
6.2未來研究方向 37
附錄 38
參考文獻 39

[1] Y. Jing, S. Baluja, “PageRank for Product Image Search,” WWW 2008, Beijing, China, 2008.
[2] M. Lew, et al., “Content-Based Multimedia Information Retrieval: State of the Art and Challenges,” ACM Transactions on Multimedia Computing, Communications, and Applications, pp. 1–19, 2006.
[3] C. Liu, Dong Wang, X. Liu, C. Wang, L. Zhang, B. Zhang, “Robust Semantic Sketch Based Specific Image Retrieval,” IEEE , 2010.
[4]C. Carson, M. Thomas, S. Belongie, J. Hellerstein, J. Malik, “Blobworld: A System for Region-Based Image Indexing and Retrieval,” Proceedings of VISUAL 99, Amsterdam, The Netherlands, pp. 509-516, 1999.
[5] F. Ban, “Content-Based Image Retrieval Using Hand-Drawn Sketches and Local Features: a Study on Visual Dissimilarity,” These n. 1312 Imprimerie St. Paul, 2000.
[6] D.P. huijsmans, S. Poles, M. Lew, “2d Pixel Trigrams for Content Based Image Retrieval,” Computer Science Department, Leiden University, 1996.
[7] D. Huijsmans, M. Lew, D. Denteneer, “Quality Measures for Interactive Image Retrieval with a Performance Evaluation of Two 3x3 Texel-Based Methods,” Proceedings of the International Conference on Image Analysis and Processing, ICIAP 97 , Firenze, Italy, pp.23-29, 1997.
[8] M. Flickner, H. Sawhney, W. Niblack, J. Ashley, Q. Huang, B. Dom, M. Gorkani, J. Hafner, D. Lee, D. Petkovic, D. Steele, P. Yanker, “Query by Image and Video Content: The QBIC System,” IEEE Computer, pp. 23-32, 1995.
[9]施政瑋，「設計資料庫系統與CBIR於圖像搜尋及管理之整合應用」，碩士論文， 國立雲林科技大學，雲林，2007。
[10] A.D. Bimbo, P. Pala, “Visual Image Retrieval by Elastic Matching of User Sketches,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 19, no. 2, 1997.
[11] 許志祥，石昭玲，「以形狀為基礎之葉子影像檢索系統」，碩士論文，中華大學，新竹，2005。
[12]葉本源，「適用於臺灣各種車輛之車牌辨識系統」，碩士論文，中原大學，桃園，2006。
[13]蕭如宣，陳彥廷，「觸控面板筆劃萃取之線上中文手寫辨識系統設計」，亞東學報第29期，台北，第25-36頁，2009。
[14] H. Sakoe, S. Chiba, “Dynamic Programming Optimization for Spoken Word Recognition,” IEEE Trans. Acoust. Speech Signal Process. Vol. ASSP-26, No. 1, pp. 43-49, 1978.
[15] Y.J. Lee, C. L. Zitnick , M.F. Cohen, “ShadowDraw: Real-Time User Guidance for Freehand Drawing,” ACM Transactions on Graphics (Proceedings of SIGGRAPH), Vancouver, Canada, 2011.