The Study of Multispectral Imagery Segmentation and Compression Technique

• Main Authors: Yao-Cheng Yang, 楊耀丞
• Other Authors: Shun-Hsyung Chang
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/36220035583544735077

碩士 === 國立海洋大學 === 電機工程學系 === 88 === Building digital libraries has become white hot in this era of internet and the World Wide Wed. For the digital library applying to the marine technology, the multispectral satellite image is essential. In additions, the applications and esearches of satellite image set become more and more important, which can be applied to the nation boundary survey, ocean resourrces developing and the detection of national defence. Concerning satellite image, their accessibility is hindered by the size of images and communication bandwidth. To alleviate these limitations, it is essential to develop a compression technique for multispectral imagery. In this paper,we propose an efficient image compression technology for multispectral imagery. Typical satellite image set exhibits a number of different terrains, and various terrains with different spectral characteristics. In order to achive better compression ratio,it is necessary to segment the image according to local terrain characteristics. In the thesis, we first propose a Eigen Region Segmentation method which segment the image into several eigen-regions under the minization of the distance between principal eigenvectors of adjacent image blocks. The simulation results show that this segmentation method is suitable for multispectral imagery. Then we develop the Eigen Region KLT compression method which combines the Eigen Region Segmentation methed with the KLT-JPEG methed. This methed transforms each eigen region by the corresponding KL transformation and produces a resulting spectrally decorrelated eigen region image. Then the eigen region images are further compressed via JPEG standard. However, the information of eigen-images via KL transformation will centralize on a few spectral bands. To further increase the compression ratio of multispectral image, we select the optimum band number for each eigen region image based on the distribution of eigenvalues.