English  |  正體中文  |  简体中文  |  Items with full text/Total items : 26988/38789
Visitors : 2345219      Online Users : 45
RC Version 4.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
Scope Adv. Search
LoginUploadHelpAboutAdminister

Please use this identifier to cite or link to this item: http://ntour.ntou.edu.tw:8080/ir/handle/987654321/11088

Title: 應用GCS神經網路於3D影像重建
Applying GCS Neural Network to 3-D Image Reconstruction
Authors: 王榮華
Contributors: NTOU:Department of Electrical Engineering
國立臺灣海洋大學:電機工程學系
Keywords: 影像重建;類神經網路;特徵抽取;立體視覺;位差
Image reconstruction;Neural network;Feature extraction;Stereo vision;Disparity
Date: 1999
Issue Date: 2011-06-28T08:08:22Z
Abstract: 本計畫研究建立一神經網路動態立體視覺系統(Active stereo vision system)。此系統除了能主動的控制攝影機之角度,使得欲觀察之區域能出現於兩攝影機之影像上,並使用不同類型神經網路來處理影像分割、完成Stereo matching的相關運算步驟,最後建立此系統可見範圍的精確深度圖。首先,為改善影像分割之效果,我們使用一個自我發展(Self-creating)學習演算法,稱做Growing cell structures(GCS)。每個神經元被賦予一個resource counter「資源計數器」,以記載其在競爭式學習中贏和輸的機率。此「資源計數器」充分能反映輸入資料之統計特性,使得網路能適時增加或移除神經元。在任何時刻,網路均維持活動力總和為一個常數,稱之為資源守恆。由於GCS嘗試最大化其熵值,我們稱GCS是一個近最佳解的向量量化器。GCS的自我產生以及可調學習率之機制使得GCS能兼顧網路的學習能力與穩定性。在影像編碼的實驗中,我們比較了GCS和三種有名演算法---Kohonen self-organizing feature map(SOFM)[14],FSCL[15]和SCONN2[16]的效能;模擬結果顯示出GCS比其他方法之優越處。 我們進一步將GCS運用在一個以神經網路為基礎的立體視覺系統中,藉由Sobel之運算得到立體影像每一點之特徵,如灰階值,變化量和方向性。再將這些特徵給GCS進行叢集影像;相同的特徵也用來訓練倒傳遞網路成為一個匹配器。訓練後的網路匹配器可用來產生一個初始的深度圖。之後,再使用一個以影像叢集為基礎的對應點演算法,以得到一個更精確的深度圖。實驗結果顯示出倒傳遞網路匹配器和對應點演算法的成效。
A self-creating neural network effective in learning vector quantization, called GCS (Growing cell structures) is introduced. Each neuron in GCS is characterized by a measure of resource. Conservation is achieved by bounding the summed resource of all neurons at a constant, despite value for which varies from one network to another. Resource values of all neurons are updated after each input presentation. We show that GCS effectively fulfills the conscience principle and achieves biologically plausible self-creating capability. In addition, conservation in resource facilitates systematic derivations of learning parameters, including the adaptive learning rate control useful in accelerating the convergence as well as in improving node-utilization. The performance of GCS is compared with three famous algorithms---Kohonen's Self-Organizing Feature Map (SOFM) [14], FSCL [15], SCONN2[16]. The GCS is further used in a stereo vision system based on neural networks [4]. Sobel operators are used to extract features of intensity, variation, and orientation from stereo image pairs. These features are used to clustering images by GCS and to train a BP neural network in order to obtain an adaptive matcher. The trained BP matcher can generate an initial or primitive disparity map that provides necessary correlation or corresponding SSD (sum of squared differences) in area-based matching methods. Following the BP training, we propose a matching algorithm based on image clustering that is useful in iteratively updating the primitive disparity map. We show that this update algorithm can improve the quality of the disparity map significantly. Empirical results show that the efficiency of the BP matcher and the validity of our matching algorithm.
URI: http://ntour.ntou.edu.tw/ir/handle/987654321/11088
Appears in Collections:[電機工程學系] 研究計畫

Files in This Item:

File Description SizeFormat
RRPB88020090.pdf376KbAdobe PDF171View/Open


All items in NTOUR are protected by copyright, with all rights reserved.

 


著作權政策宣告: 本網站之內容為國立臺灣海洋大學所收錄之機構典藏,無償提供學術研究與公眾教育等公益性使用,請合理使用本網站之內容,以尊重著作權人之權益。
網站維護: 海大圖資處 圖書系統組
DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - Feedback