A modified SIFT descriptor for image matching under spectral variations

Abstract

In multispectral imaging multiple discrete wavelength bands are used to image a scene. The imaging process maps the scene contents to different intensity levels and varies the scene appearance from band to band. This induces intensity variations among the spectral images and effects the performance of SIFT for cross spectral image matching. This paper proposes modifications to the SIFT descriptor in order to improve its robustness against spectral variations. The proposed modifications are based on fact, that edges remain well preserved in multispectral imaging and we can achieve better image matching results by boosting the contribution of local edges in the SIFT descriptor construction process. Therefore, we propose a Local Contrast (Δ) and a Differential Excitation (ξ) function for the construction of SIFT descriptors. The experimental results show, that the performance of Δ-SIFT and ξ-SIFT is superior to standard SIFT for image matching under spectral variations. © 2013 Springer-Verlag.