Spatial Resolution Enhancement of Remote Sensing Hyperspectral Images With Localized Spatial-Spectral Dictionary Pair

• Main Authors: Yifan Zhang, Jin Tian, Tuo Zhao, Shaohui Mei
• Format: Article
• Language: English
• Published: IEEE 2020-01-01
• Series: IEEE Access
• Subjects: Collaborative representation; dictionary; hyperspectral; image fusion; multispectral; resolution enhancement
• Online Access: https://ieeexplore.ieee.org/document/9040549/

The hyperspectral image (HSI) is capable of providing abundant and detailed spectral information in hundreds of contiguous spectral bands. While due to some practical reasons, its spatial resolution is generally lower than that of multispectral image (MSI) and panchromatic image. To deal with the limited spatial resolution issue of HSI, a low resolution (LR) HSI can be fused with a high resolution (HR) MSI of the same scene to generate an HR HSI. A novel dictionary-based HSI and MSI fusion method (SSLDF method) is proposed in this paper, in which a localized spatial-spectral dictionary pair incorporating both spatial and spectral information simultaneously is constructed and adopted, rather than the traditional spectral or spatial one. To construct the HR and LR dictionary pair, HR MSI and its spatial degradation (LR MSI) are divided into overlapped subimages. Furthermore, to reduce the dictionary scale and hence to efficiently reduce the computation cost, a localized strategy is employed for dictionary construction rather than a global one, which makes atoms of the spatial-spectral dictionary actually all patches within the subimage. Based on the appropriate assumption that the LR HSI and HR HSI (expected fusion result) can be collaboratively represented by LR dictionary and HR dictionary respectively sharing the same set of representation coefficients, the desired HR HSI is reconstructed by HR dictionary and the collaborative representation coefficients obtained with LR HSI and LR dictionary. In simulative experiments, the newly proposed SSLDF method is validated and compared with both state-of-the-art dictionary-based fusion methods and representative fusion methods not limited to the dictionary-based ones. Simulative experimental results illustrate that the proposed fusion method is capable of producing better or comparable fused results compared with these representative fusion methods. Its simple structure as well as low computation cost makes it quite promising in practical applications.