一种基于张量字典学习的光谱图像稀疏编码方法

A sparse coding method for spectral images based on tensor dictionary learning

光谱图像与传统RGB图像不同,光谱图像包含有连续的光谱曲线,这些信息对目标识别,目标分类等领域有着很高的应用价值。但是当光谱分辨率较高时,光谱图像数据量剧增,将占用更多的存储空间,为光谱图像的储存和传输带来了难度。因此,本文提出了一种基于张量字典学习的光谱图像稀疏编码方法,主要通过将光谱图像以张量的形式进行表示,增强数据之间的空间相关性;在保证图像不失真的前提下使用先验的光谱字典对光谱图像进行稀疏编码,通过储存该稀疏编码代替原始数据,从而减低降低光谱图像储存量。

Different from traditional RGB images,spectral images contain additional continuous spectral curves,these information plays an important role in target recognition,target classification and other fields.However,when the spectral resolution is high,the amount of spectral image data increases dramatically and will take up more storage space.This makes it difficult to store and transmit spectral images.In this paper,a sparse coding method for spectral images based on tensor dictionary learning is proposed.By representing spectral images in tensor form,the spatial correlation between data is enhanced.Under the premise of ensuring that the image is not distorted,a priori spectral dictionary is used to perform sparse coding on spectral images.By storing this sparse coding instead of the original data,the storage capacity of spectral images is reduced.