利用复合变换的多光谱图像压缩算法  

Multispectral image compression algorithm based on composite transform

作　　者：梁玮[1] 曾平[1,2] 郑海红[1] 罗雪梅[1] 

机构地区：[1]西安电子科技大学计算机学院,陕西西安710071 [2]西安石油大学计算机学院,陕西西安710065

出　　处：《西安电子科技大学学报》2015年第4期33-40,共8页Journal of Xidian University

基　　金：国家"十二五"预研资助项目(513160702);中央高校基本科研业务费专项资金资助项目(K5051303013;K5051303014)

摘　　要：针对现有多光谱图像压缩算法去除空谱冗余不充分、自适应性不强等问题,提出了多光谱图像的空间稀疏等价表示方式以及相应的聚类实现算法——OptimalLeaders.在此基础上,设计了一种基于复合变换的多光谱图像压缩算法——OLPKWS.该算法使用OptimalLeaders聚类,将多光谱数据等价变换为代表光谱和差别成分,自适应去除其空间冗余;采用误差补偿机制提高多光谱图像重建质量;针对代表数据,采用主成分分析降维去除谱间冗余;针对预测差别成分,通过Karhunen-Loeve变换(KLT)去除谱间冗余,用二维小波变换去除空间冗余,最后采用标准差码率预分配策略结合SPIHT算法完成压缩编码.实验表明,在相同的压缩比下,所提算法较聚类、SPIHT和KLTSPIHTTCIRA算法明显提高了重建图像的峰值信噪比.Aiming at the problems of inadequate reducing of spatial and spectral redundancy and weak adaptability of existing multispectral image compression algorithms , the multispectral images＆#39; spatial sparse equivalent representation and its clustering implementation named the OptimalLeaders are proposed . Furthermore , an adaptive multispectral image compression algorithm — OLPKWS is designed , which is based on composite transform . In the OLPKWS , multispectral data are transformed into representation and residual by the presented spatial sparse equivalent transform , which removes spatial redundancy adaptively . Moreover , an error compensation mechanism is introduced in order to improve the quality of the reconstruction image . Principal Component Analysis （ PCA ） is used to remove spectral redundancy for representation . However , to predict differences , KLT is utilized to explore spectral correlation , two‐dimensional wavelet transform is used to remove spatial redundancy , standard deviation weighted rate allocation and SPIHT are combined to complete the coding . Experimental results show that , in comparison to the clustering , SPIHT and KLT_SPIHT_TCIRA algorithms , the proposed approach achieves a higher peak signal to noise ratio （ PSNR） under the same compression ratio .

关 键 词：多光谱图像压缩 空间稀疏等价表示 聚类 主成分分析 小波编码 误差补偿 码率分配 

分 类 号：TN911.74[电子电信—通信与信息系统]