【作者】 孙子翼；

【导师】 吴一全；

【作者基本信息】 南京航空航天大学 ， 信号与信息处理， 2010， 硕士

【摘要】 多光谱遥感图像在航空航天、地质勘探、环境监测等领域都有广阔的应用前景。多光谱遥感图像作为一种三维图像提供了关于地物更细致的光谱信息,但同时其数据量急剧增加,给在有限信道上传输和存储带来了困难。所以研究多光谱遥感图像的高质量的压缩技术具有十分重要的实际意义。本文主要研究基于Contourlet变换和粒子群的多光谱遥感图像分形压缩方法,包括以下几个方面:首先,实现了基于粒子群和K-均值聚类的图像分形压缩算法,进而给出了基于免疫粒子群和K-均值聚类的Contourlet域图像分形压缩算法。通过免疫粒子群和K-均值聚类算法对Contourlet变换后的系数进行分类,完成分形压缩。实验结果表明,与基于K-均值聚类的快速分形编码算法相比,本文给出的两个算法在相同压缩比下能较好地保持图像的纹理;然后,对多光谱遥感图像每个波段进行空域Contourlet变换,将Krawtchouk矩不变量和模糊C-均值聚类算法引入到分形压缩中,分别实现了基于Krawtchouk矩不变量及基于Krawtchouk矩不变量和Contourlet变换的多光谱图像分形压缩算法。实验结果表明,通过进一步对Range块和Domain块细分类,本文实现的算法能够获得更高的峰值信噪比和压缩比;接着,实现了基于核模糊聚类及基于改进粒子群核模糊聚类的Contourlet域多光谱遥感图像分形压缩算法,实验结果显示本文算法的编码时间较基于小波域的遥感图像分形压缩算法大为减少,同时获得了更好的解码图像效果;最后,通过支持向量回归逼近Contourlet系数和分形编码,并利用粒子群算法优化支持向量回归机参数,实现了基于支持向量回归的Contourlet域多光谱遥感图像分形压缩算法。实验结果表明,相对于基于支持向量机的遥感图像压缩算法,本文所给出的算法能在更高的压缩比下获得质量更好的解码图像。更多还原

【Abstract】 Multispectral remote sensing images have wide potential application, including aerospace, mineral exploration, environment monitoring, etc. As 3-D images, multispectral remote sensing images provide more precise spectral information of landmark, but results in large size data sizes. The storage and transmission of large volumes of multispectral data have become significant concerns. Therefore the research on compression methods of multispectral remote sensing images is very important and practical. This thesis contributes to research compression methods of the multispectral remote sensing images based on contourlet transform, fractal compression and PSO. The thesis mainly includes the following aspects.Firstly, the image fractal compression algorithm based on PSO and K-means is introduced. Then the algorithm based on contourlet transform, IPSO and K-means is realized. By classifying the contourlet coefficients, the experimental results show that, the realized algorithms can get better quality of image with the same compression ratio.Then, by decomposing the multispectral remote sensing image based on contourlet transform in the dimensional domain, the Krawtchouk moments and FCM is used in fractal compression. The multispectral image fractal compression algorithm based on Krawtchouk moments is introduced. Then the algorithm based on Krawtchouk moments, FCM and contourlet transform is realized. By classifying the range and domain blocks, the presented algorithms can reach higher PSNR and compression ratio.Next, to reach the better quality of reconstructed image, the algorithm of multispectral image fractal compression based on KFCM is presented. Then the method based on mtsPSO and KFCM is realized. Experimental results show that compared to the method based on wavelet transform and fractal compression, the introduced algorithms can reduce encoding time efficiently and reach better quality of image.Finally, SVR is used to approximate the contourlet coefficients and fractal codes. Then the SVM parameters are optimized by PSO. The multispectral remote sensing image fractal compression method based on SVR and contourlet transform is proposed. The experimental results show that, compared to the method based on SVM, the realized method can reach higher PSNR in similar compression ratio.更多还原