歐陽明臺灣大學:資訊工程學研究所高弘政Kao, Hong-ChengHong-ChengKao2010-05-182018-07-052010-05-182018-07-052008U0001-1206200814114500http://ntur.lib.ntu.edu.tw//handle/246246/183610數位相機的普及率越來越高，因此越來越多人拍照，使得數位照片成數性成長。面對龐大數量的照片，照片的整理與挑選成了一個很大問題。片構圖的意思是相機位置的擺設以及視野的選擇，決定什麼物體放進來或該排除在外，一般來說，照片構圖幾乎決定了一張照片的壞，所以我們針對照片構圖的部分做研究。照片構圖這個問題往往主觀的，也跟人類的視覺感受有關，雖然是主觀的，但幸運的是，據攝影學家多年來的經驗，可以萃取出一些比較通用性的原則，這規則在各種談攝影的書中被談到。我們從書中挑選出幾個定義較明、且較適用於電腦量化分析的規則，將他們實作成可以自動化執行程序，用這些規則挑選出比較不好的照片，以及對照片做一些建議評分。張不是很好的照片如果經過很好的裁切，可以讓一張照片起死生，現在數位相機的能力也越來越強，照片的解析度越來越高，在有輸出太大尺寸照片需求的原則下，很有本錢可以對照片做一些裁，我們以構圖分析的結果來做裁切的原則，有機會能讓照片變的更。們初步的結果是基於132 張照片，這些照片是針對每個規則（水線、照片平衡、主體位置、線條及形狀、避免融合）挑選出比較化的情況，每個規則的準確率分別在71%、96.8%、73.1%、78.4%以71.4%。Digital camera is very popular and the number of digital photos grows exponentially.aced to huge number of photos, the collection and selection ofhotos becomes a big problem.hoto composition means the placement of camera and the selection ofhe field of view. It determines whether objects should be placed inside thehoto or be excluded outside. Most people agree that photo compositionlmost determines whether a photo is good or not. Therefore, our research isocus on photo composition. This problem is subjective and relates to humanisual perception. Although this problem is subjective, it is fortunate thatrom the experience accumulated by photographers in recent years, certainommon rules were extracted. We select the rules that have clear definitionnd suitable for automatic quantitative analysis to pick out bad photos or toake recommendations and scores.t is said that a not very good photo after doing a good cropping will let ahoto turn from death into life. The performance of digital camera nowadayss more and more powerful and image resolution is higher and higher, ando images can be cropped without serious loss of resolution. Our results inutomatic cropping support this observation.ur preliminary results for 132 photos are based on five rules (the horizon,hoto balance, location of main object, revealing line patterns and shapes,nd avoiding mergers), and the precision of each rule is 71%, 96.8%, 73.1%,8.4%, and 71.4%, respectively.致謝i要iibstract iii Introduction 1 Related work 4.1 Recomposition, information preserving, and automatic cropping . . . . . 4.2 Feature extraction tools for photo composition . . . . . . . . . . . . . . . 5.2.1 Canny edge detection . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Hough transform for line detection . . . . . . . . . . . . . . . . . 9.2.3 Region of interest (ROI) . . . . . . . . . . . . . . . . . . . . . . 11.2.4 Face detection . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Rules of esthetics in photo composition 17.1 The horizon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.2 Photo balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18.3 Location of main object: the rule of thirds . . . . . . . . . . . . . . . . . 18.4 Revealing line patterns and shapes . . . . . . . . . . . . . . . . . . . . . 18.5 Avoiding mergers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.6 Others: Reserving more space for the viewing direction, etc. . . . . . . . 20 Implementation 22.1 Scoring system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.1.1 The horizon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.1.2 Photo balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.1.3 Location of main object . . . . . . . . . . . . . . . . . . . . . . 25.1.4 Revealing line patterns and shapes . . . . . . . . . . . . . . . . . 26.1.5 Avoiding mergers . . . . . . . . . . . . . . . . . . . . . . . . . . 29.1.6 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.2 Automatic correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.2.1 Rotational correction . . . . . . . . . . . . . . . . . . . . . . . . 30.2.2 Cropping correction . . . . . . . . . . . . . . . . . . . . . . . . 32 Results 33.1 The horizon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33.2 Photo balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35.3 Location of main object . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.4 Revealing line patterns and shapes . . . . . . . . . . . . . . . . . . . . . 37.5 Avoiding mergers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37.6 Better photo composition by cropping . . . . . . . . . . . . . . . . . . . 37 Conclusion and future work 40.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41ibliography 44application/pdf34194342 bytesapplication/pdfen-US照片構圖美學規則照片平衡三等分原則感興趣區域照片裁切Photo compositionesthetics rulesphoto balancethe rule of thirdsregion of interest (ROI)photo croppingthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/183610/1/ntu-97-R95922060-1.pdf