摘要
高光谱图像的稀疏分解能得到其稀疏表示形式,便于对图像进行压缩处理。因高光谱图像特征复杂,单一正交基无法捕捉到图像信号的所有特征,需构建原子个数更多的冗余字典对高光谱图像进行稀疏表示。针对高光谱图像,以高斯原子为基础,构造三种冗余字典,利用正交匹配追踪算法找到最优原子,完成高光谱图像的稀疏分解,利用重构图像的峰值信噪比、结构相似性和计算效率对冗余字典的稀疏表示能力进行评价。实验结果表明,构造的三种冗余字典均能对高光谱图像进行稀疏表示,重构图像的峰值信噪比均能达到40dB以上,结构相似性达到0.99以上,且高斯字典的计算效率最高。
Sparse decomposition of hyperspectral image could obtain the sparse representation form of image,which could facilitate image compression. Due to the features of hyperspectral images are complex,the single orthogonal basis could not capture all the features of the images. Therefore,the redundant dictionary contained more atoms is constructed to sparse represent the hyperspectral images. For hyperspectral images,based on the Gaussian atom,three kinds of redundant dictionaries are proposed in this paper. Orthogonal matching pursuit algorithm is utilized to seek for the optimal atoms to achieve the sparse representation process of hyperspectral images. The reconstructed peak signal-to-noise ratio, structural similarity, and computational efficiency are used to evaluate the sparse representation ability of the three dictionaries. Experimental results show that all the constructed dictionaries could sparse represent the hyperspectral images very well,and the reconstructed peak signal-to-noise ratio could hit 40 dB or more,the structural similarity could reach 0.99 or more,and the computational efficiency of Gauss dictionary is the highest.
作者
王丽
王威
WANG Li;WANG Wei(School of Electronic Engineering,Xi’an Aeronautical University,Xi’an 710077,China)
出处
《电子设计工程》
2019年第10期107-112,共6页
Electronic Design Engineering
基金
西安航空学院校级科研基金(2018KY1222
2016KY1206)
关键词
稀疏分解
冗余字典
正交匹配追踪
峰值信噪比
sparse decomposition
redundant dictionary
orthogonal matching pursuit
peak signal-to-noise ratio
