English  |  正體中文  |  简体中文  |  Items with full text/Total items : 26988/38789
Visitors : 2318017      Online Users : 29
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/987654321/34422

Title: 針對高頻譜影像以階層式影像切割為基礎之聚類分群壓縮法研究
Group-Based Compression Method Using a Hierarchical Segmentation Approach for Hyperspectral Imagery
Authors: 張麗娜;張陽郎
Contributors: NTOU:Department of Communications Navigation and Control Engineering
國立臺灣海洋大學:通訊與導航工程學系
Keywords: 高頻譜影像;階層式影像切割;影像壓縮;四元素矩量保持法;主成份分析;位元分配
hyperspectral images;hierarchical image segmentation;image compression;binary quaternion-moment-preserving(BQMP) thresholding technique;Principal Components Analysis(PCA);bit allocation
Date: 2012
Issue Date: 2013-10-07T02:31:00Z
Abstract: 高頻譜影像有更多光譜波段數目和更高的空間解析度,相對的比多頻譜影像提供 更精確和豐富的資訊與辨識能力。然而高解析度意味著高資料量,除了增加影像處理 困難度也造成影像資料傳輸的效能不佳。考慮高頻譜影像在頻譜和空間存在著高資料 相關性,即存在著大量的冗餘性的資料,因此在有限的通訊頻寬下,發展一高效率且 適用於高頻譜影像的壓縮技術是一關鍵性的研究課題。本計畫擬充份利用高頻譜影像 頻帶間和空間資料的相關性,提出結合階層式影像切割和頻帶聚類分群技術之影像壓 縮法。研究中擬先提出一以四元素矩量保持法為基礎之階層式影像切割技術,此法可 將影像依其內容將具高空間相關性資料聚集為相同的影像區間。接著,擬對各影像區 間以頻帶聚類分群法將具高頻域相關性的光譜波段聚為同一群組,藉此程序可使高空 間和頻域相關性資料聚集為相同的群組,將有利於資料冗餘性的移除。最後,再對各 群組以主成份分析法結合JPEG2000 進行頻域和空間域影像資料壓縮。本研究將於兩年 內完成,第一年將以階層式影像切割為基礎提出聚類分群壓縮法;第二年將針對聚類 分群壓縮法關鍵模組進行優化,以提升聚類分群壓縮法的效能。 本計畫為期二年,研究重點包含下列各項: 第一年:以階層式影像切割為基礎之聚類分群壓縮法研究 (1) 簡化萃取高頻譜影像特徵的研究:此部分將影像先依頻帶間的相關性分為群組, 再對各頻帶群組萃取其特徵,藉此減少直接萃取高頻譜影像特徵的計算複雜度並 達到降低資料維度的目的。 (2) 階層式影像切割的研究:擬依(1)萃取的特徵以遞迴方式進行階層式分割,此程序 可將影像依其內容切割為適當的影像區間。 (3) 聚類分群壓縮法之研究:以影像區間為基礎發展頻帶分群的影像壓縮法,將(2)影 像區間的頻帶依光譜波段間的相關性分為群組,再針對每一群組進行對應的主成 份轉換和JPEG2000 影像壓縮,以加強壓縮效能。 (4) 電腦模擬及效能評估:以電腦模擬驗證階層式影像切割和壓縮法的效能,並提出 後續改善方向。 第二年:提升聚類分群壓縮演算法效能之研究 (1) 主成份轉換主要數目決定之研究:利用訊號自相關矩陣和共變異矩陣對應的特徵 值具一致性分佈的特性,以決策理論決定主要成分數目,改善以訊號能量決定主 成分數目時易忽略少數卻重要特徵的缺陷。 (2) 位元分配之研究:各影像群組經主成份轉換後,依所含的訊息量分配進行JPEG2000 資料壓縮的位元數,以提升壓縮效能。 (3) 資料傳輸格式之研究:聚類分群壓縮法以不規則影像區間為單位,傳統衛星影像 的格式已不適用,擬提出對應的資料傳輸格式以適用於on-board 衛星之應用。 (4) 電腦模擬與效能評估:以電腦模擬驗證壓縮法的效能,並評估此法的計算複雜度 及討論實現此壓縮法的可能性。
Hyperspectral imagery with hundreds of bands offers high spectral resolution and provides the potential accuracy in detection and classification of targets unresolved in multispectral images. However, higher spectral resolution increases the computation complexity in image processing. Thus, how to effectively deal with its enormous data volume is the main challenge for hyperspectral image analysis. Concerning the hyperspectral images, their accessibility is hindered by the size of image and communication bandwidth. To alleviate this limitation, it is essential to develop a high efficient compression technique for hyperspectral images. Since hyperspectral images reveal high degree of spectral and spatial correlations, in the studies, we prepare to exploit the properties, and develop an efficient group based compression method which combines a hierarchical segmentation technique and a band clustering approach. At first, we prepare to propose a hierarchical image segmentation which combines the grouped-based image feature extraction and a binary quaternion-moment-preserving (BQMP) thresholding technique. Then, the spectral bands are clustered into several groups according to their associated band correlation for each image region. Thus, image data with high degree correlations in spatial/spectral domains are then gathered into groups. Finally, the grouped image data is compressed by one-dimensional (spectral) Principal Components Analysis (PCA) and two-dimensional (spatial) JPEG2000 compression techniques. The project will last two years. In the first year, we will develop a group-based image compression technique. In the second year, we will do fine tune of key subjects of group-based image compression such as determining the number of principal components in PCA and a bit allocation rule for JPEG 2000 in order to improve the compression efficiency,. The proposed methods will provide a more accurate segmentation result and improve the compression efficiency of hyperspectral images with less computation complexity. The project will last two years and our studies are mainly in the following aspects: 1. In the first year: A group-based compression method using a hierarchical segmentation approach for hyperspectral imagery  Extract image features for each band group by PCA.  Develop a hierarchical image segmentation by using BQMP thresholding technique to partition the hyperspectral images into proper regions.  Partition the bands into groups for each region and derive a group-based compression method which performs PCA transform and JPEG2000 for each image group.  Evaluate the performance of the proposed approach. 2. In the second year: Enhancing performance of the group-based compression method  Determine the number of principal components in PCA.  Develop a bit allocation rule for each image group before compressed by JPEG 2000..  Propose a proper transmission format for the compression method.  Evaluate the performance of the proposed approach.
URI: http://ntour.ntou.edu.tw/handle/987654321/34422
Appears in Collections:[通訊與導航工程學系] 研究計畫

Files in This Item:

There are no files associated with this item.



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