臺灣博碩士論文加值系統

在規則排列的圓形中，觀察每個圓形，可以發現其邊緣是由上、下、左、右四個相等大小之弧形構成。而每一排的圓形其上方弧形及下方弧形分別呈共線狀態(可排成一列虛線)，而左方及右方的弧形亦然。
本論文主旨是藉由霍夫轉換偵測影像中規律分布之圓形，其動機源自於觀察規律排列之圓形，其影像經由邊緣偵測後會有許多相同方向之片段結構，這些片段結構會連成一條相同方向之虛線。本論文亦會對各種規律排列的非圓影像進行測試。
偵測流程分成五個步驟，如下所示：
(1) 以邊緣偵測擷取影像中水平及垂直結構，取出兩組不同角度且包含圓弧分布的邊緣線段結構圖。
(2) 兩組角度分別對應到霍夫空間後，在霍夫空間中利用閥值將可能為弧形所連成之直線代表點找出來。
(3) 接著將這些在霍夫空間的點，對應回原影像之直線，就各直線所經之邊緣線段之長短判斷後捨棄非目標之過長線段。
(4) 接著利用霍夫轉換對兩組結構圖執行第三步驟之處理，直到去除所有非目標線段。
(5) 合併兩種處理過之結構圖並標記出圓形可能的所在區塊。
在本研究中，對於有較高對比及較近距離的斑點或圓型服飾通常都可以偵測出來，但對於中間對比度或相對低對比度、較長線段及城市中之建築物並無法有效的偵測出來。

Consider the case where a circular boundary consists of 4 equal-sized arcs, the upper/lower and left/right ones, derived from an array of circular disks lying horizontally in equal distance. It’s readily observed that all upper-arcs “lie” in the same line, as “lie” all lower-arcs in another. The same phenomenon exists for left/right arcs on vertically distributed disks.
The thesis addresses the subject of detecting imagery patterns of somewhat regularly distributed circular disks via Hough transform. The conception arises from the observation that the oriented-fragments in an edge-map derived from an image of circular disks in quasi-regular formation appear to lie on a same line in the very same orientation. A prototype system was setup for a series of detecting task in which quite some images of repetitive patterns, besides circles, are examined.
The detecting process takes five steps as follows.
(1) Extraction of horizontal and vertical arcs from disks in the input image, resulting in 2 arc maps;
(2) Derivation of counterparts in Hough-space of the 2 arc maps, followed by proper threshold;
(3) Inverse-transform of the 2 Hough-space maps for lines of arc-in-line in the 2 arc maps, followed by the discard of “flawed-arcs” which are unlikely to be parts of circles;
(4) Hough-transform of the 2 refined arc maps in (3), followed by Inverse Hough-transform and subsequent flaw-arc removal as done in (3);
(5) Congregation of the 2 refined arc maps for identifying the image regions where quasi-circular patterns exist.
The initial results shown by the pilot system are deemed as satisfactory for images with spotted or circular patterns on people’s clothing where high contrast and tight spacing is generally found. For repetitive patterns accompanied with medium or relatively low contrast and in relatively large distance, as already being encountered in images of buildings in urban scene, a similar line of process as deployed in the prototype could be applied, though not yet implemented.

chap1 緒論 1
1.1 簡介 1
1.2 論文架構 3
chap2 實驗動機及相關研究 4
2.1 實驗動機 4
2.2 相關研究 5
chap3 處理流程 10
3.1 簡介與系統流程架構 10
3.2 水平、垂直結構偵測 12
3.3 霍夫轉換 14
3.3.1 霍夫轉換介紹 14
3.3.2 本研究之霍夫空間介紹 16
3.4 線段選取及整合 18
3.4.1 偵測角度限制 18
3.4.2 局部極大值 19
3.4.3 忽略長線段之誤判 20
3.4.4 單一角度選擇 21
3.4.5 線段分群及整合 22
3.5 結構圖之改善 27
3.5.1 線段結構的判定 28
3.5.2 長線段結構的移除 30
3.5.3 合併兩種結構 33
3.6 改善後之實驗結果 35
chap4 實驗結果 38
4.1 簡介 38
4.2 規律排列之圓形 39
4.2.1 相距較遠之小圓 39
4.2.2 相距較近之小圓 50
4.2.3 大圓 58
4.3 規律排列之非圓形 68
4.3.1 碎花 68
4.3.2 格紋 74
4.3.3 建築物 83
chap5 結論 89
文獻參考 91

[1] DUDA, Richard O.; HART, Peter E. Use of the Hough transformation to detect lines and curves in pictures. Communications of the ACM, 1972, 15.1: 11-15.
[2] CHIU, Shih‐Hsuan; LIAW, Jiun‐Jian. A proposed circle/circular arc detection method using the modified randomized Hough transform. Journal of the Chinese Institute of Engineers, 2006, 29.3: 533-538.
[3] DOUBEK, Petr, et al. Image matching and retrieval by repetitive patterns. In: Pattern Recognition (ICPR), 2010 20th International Conference on. IEEE, 2010. p. 3195-3198.
[4] SOBEL, Irwin; FELDMAN, Gary. A 3x3 isotropic gradient operator for image processing. a talk at the Stanford Artificial Project in, 1968, 271-272.
[5] 陳慶瀚，" Ch8 Hough Transform"，from I-Shou University, Web site: http://www2.nkfust.edu.tw/~wenh/2008/multimedia/slides/Hough.pdf
[6] 邱承縉、林正中，"都會區建築物偵測"，碩士論文，2013.
[7] 陳至名、林正中，"鉛筆素描模擬中粗疏淡化筆觸於非描繪重點區域之引用與相關呈現處理"，碩士論文，2014.
[8] 薛智仁、林正中，"視覺對稱區域之梯度分佈結構"，碩士論文，2012.
[9] 何明鴻、林正中，"透過梯度資訊偵測影像主體之對稱性"，碩士論文，2011.