- 學位論文
結合顏色、紋理及外型為特徵的影像檢索系統
A Content-Based Image Retrieval System based on Color, Texture and Shape Features
摘要
影像檢索技術(Content-Based Image Retrieval, CBIR)已經引起了廣泛學者的研究，有別於一般以文字檢索的方式，影像檢索籍由輸入影像來檢索相似或相關聯的影像，最後透過手機、網站或其它媒體呈現給使用者。 一開始本論文將提出一種使用顏色及紋理作為影像特徵的影像檢索系統(Color and Texture Image Retrieval System, CTIRS)，此系統使用自組織神經網路映射圖(Self-Organizing Map)作為顏色特徵，定型樣式並行出現矩陣(Motif Co-occurrence Matrix)作為紋理特徵，實驗證明，此系統不但能有效的檢索出顏色相近的圖片外，還能迅速的比對出具有相似紋理的影像，從實驗結果顯示，此系統在大部分類別的影像都具有相當高檢索的準確率，但對於部分顏色及紋理特徵不明顯的影像，準確率尚有加強的空間。 因此本論文另外又提出了一種結合顏色、紋理及外型的影像檢索系統(Color, Texture and Shape Image Retrieval System, CTSIRS)，其大致架構與特徵與前項系統相似，只是加入了一種基於外型屬性的影像特徵物件力矩(Object-Moment)，並結合特徵選取的機制，籍此來提高系統之效能。從最後實驗結果得知，CTSIRS可以有效的提昇檢索之準確率。
並列摘要
Content-based image retrieval (CBIR) has been studied by many researchers, differ from other traditional text-based retrieval technique. CBIR search image by input similar image and display through the cell phone or web. In the beginning of this article, we will propose an image retrieval system which combining color and texture as image feature call Color and Texture Image Retrieval System (CTIRS). This system using Self-Organizing Map as the basement of color feature and Motif Co-occurrence Matrix as the basement of texture feature. That in most of cases the system performance is very well except some image which image object has similar shape. In order to increase the system ability to handle more images, we will propose another image retrieval system Color, Texture and Shape Image Retrieval System (CTSIRS). The different of the CTSIRS and CTIRS is CTSIRS add a shape-based image feature Object-Moment (OM). OM is based on a simple concept the moment of force to represent the object edge in image. In addition, in order to increase the system performance, feature selection technique Sequential Forward Selection (SFS) has also been adopted to select suitable feature set. The experiment result shows that CTSIRS can efficient increase the retrieval precision.