Towards automated materials analysis: multi-scale spatio-spectral approach for registering secondary electron hyperspectral images

Secondary electron hyperspectral imaging (SEHI) is an emerging technique that allows experimental investigation of the chemical properties of material surfaces at the nanoscale. Despite advancements in automated image registration, registering SEHI images across different spatial scales remains a challenging task and it is still typically performed manually in practice. Many image registration approaches primarily focus on spatial features and they do not capture the rich spectral information in SEHI, often leading to inadequate registration. To fill in this gap, we propose a multi-scale spatio-spectral approach based on the scale-invariant feature transform (SIFT) to automate SEHI image registration. The main novelty of this work lies in the proposed multi-scale SIFT spatio-spectral descriptors that effectively integrate both SIFT spatial descriptor and differentiate the spectral signal profile as a function of electron energy. A new spatio-spectral descriptor matching algorithm is designed, achieving accurate image registration, with an accuracy error of less than 0.44 pixels. Results over real SEHI datasets how that the proposed approach performs better than other state of-the-art registration methods. Two performance metrics are used for evaluation, including normalized cross correlation (NCC) for examining spatial alignment and cosine similarity regarding spectral alignment. The results show that our proposed approach achieves good accuracy, with an NCC of above 82%, and a cosine similarity of over 93%. This work offers a promising solution for automated multi-scale image registration, which is an essential step for SEHI analysis tasks for material surface chemical composition and accelerating materials discovery.