請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/41063
標題: | 應用於學習和追尋影像內涵之意向視注覺模型 Intention-oriented visual attention model for learning and seeking image content |
作者: | Yu-Wei Huang 黃昱瑋 |
指導教授: | 林巍聳(Wei-Song Lin) |
關鍵字: | 視覺,視注覺,類神經網路,影像蒐尋,模糊邏輯, machine vision,visual attention,neural network,image content searching,fuzzy logic, |
出版年 : | 2008 |
學位: | 碩士 |
摘要: | 人的視覺可以根據意向在視野中尋找特定的目標,實質上這是一個結合影像訊息和心理意願的視覺智能,而意向式視注覺演算模型就是實現這個視覺智能的具體方法。本研究的目的是發展具有學習能力的意向式視注覺演算模型,讓意向可以用影像樣本表示,而系統可以自動從樣本萃取代表意向的模糊邏輯規則,做為蒐尋特定目標的依據。有學習能力的意向式視注覺演算模型會從影像中萃取亮度、色澤、形狀等特徵,再用中央與周圍對比機制篩選特徵突顯的區塊,模糊類神經網路則會根據代表意向的模糊邏輯規則,從突顯的區塊中選出符合意向的標的;意向式視注覺演算模型是仿效既有的方法,而有學習能力的意向式視注覺演算模型則是本研究的主要貢獻,使系統可以自動從影像特徵萃取代表意向的模糊邏輯規則,並用以蒐尋符合意向的目標。有學習能力的意向式視注覺演算模型可應用於影像內容蒐尋,本研究以輔助駕駛人蒐尋路標目標的實驗展示其可行性。 Human vision can intentionally find a target in the view. Technologically, this ability is a data- and intention-driven mechanism, and the intention oriented computational visual attention (ICVA) model attempts to imitate it by computational intelligence. This research contributes to enabling the ICVA model with learning ability so as to acquire or change intention according to assigned image samples. This innovative design is called the self-learning ICVA model which contains a neuro-fuzzy network to learn intention from image samples. A well-trained self-learning ICVA model can find interested objects in images by extracting attentive areas and matching them with intention expressed by fuzzy rules. Image features are extracted from intensity, color hue, shape and symmetry channels. Attentive areas are determined by implementing the center-surround mechanism. Intention or interested image content is assigned by image samples. By extracting fuzzy rules from image samples, the self-learning ICVA model acquires or changes the intention. The whole design is verified by constructing an intelligent road sign detection system. Experimental results show the system succeeds in learning and seeking image content with rectangular road signs. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/41063 |
全文授權: | 有償授權 |
顯示於系所單位: | 電機工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-97-1.pdf 目前未授權公開取用 | 2.86 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。