標題: 部份影像辨識之研究
A Study on Partial Image Recognition
作者: 蘇俊嘉
Su, Jiunn-Jia
林昇甫
Lin Sheng-Fuu
電控工程研究所
關鍵字: 部份影像;partial image
公開日期: 1996
摘要: 我們在此論文中,提出一種新的架構和方法,能夠正確而有效地辨識 出部份影像。為了要測試系統的性能,我們從原始影像之中隨機選取一部 份的影像,並加上作旋轉、放大或縮小的處理。在我們的系統中,我們並 不要求影像需要有固定的大小,而且也不 要求影像中需要有任何完整定 義物體的結構(well-defined structure),或存在有特定點(interest nodes),才能用來做辨識。為了能夠正確而有效地辨識部份影像,我們所 提出的方法和步驟,就是利用結合特徵偵測 (feature detection),特徵 點的直方圖 (feature nodes histogram),影像點的資料運算(node data operations)},與檢測法 等技巧,使部份影像的辨識系統能夠更一般化 和強鍵化 (generally and robustly)。 在我們的系統裡,先利用 Gabor函數運算法則抽取影像的特徵點,並以此建立資料庫 。在辨識的過 程之中,首先偵測部份影像上的特徵點,然後建立特徵點的直方圖,以便 在辨識時,利用它來決定在部份影像上的特徵點辨識時的順序。接下來, 我們再利用影 像點的資料運算(node data operations)和資料庫的資料 作比對。一但系統找到原始影 像的可能相同部份,我們再利用檢測法來 檢查所辨識的部份是否正確。經過模擬之後,顯示我們的系統有不錯的效 果。 In this thesis, we propose a new, partial image recognition system. To testour system, we process partial images at random locations, rotations, andscales from raw images. In our system, we do not limit the raw image to afixed size. And we do not need constraints on the image having any well-defined structures or any interesting node within it, either. A method for image- feature detection using Gabor functions is also described. We propose a new model that uses several algorithms and steps. That is, by combining feature detection, feature nodes histograms, node data operations, and testing rule, we have developed a system that performs generally and robustly. Our system detects feature nodes in the image by Gabor functions algorithm,then constructs a database. After detecting the feature nodes, and constructing a feature nodes histogram to sort the order of these feature node sequence, we use node data operations for matching. When possibly parts of raw images are recognized, we use a testing rule to check whether the recognized part are correct. Experimental results show that our system has good performance.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT850327001
http://hdl.handle.net/11536/61652
Appears in Collections:Thesis