Speeded-Up Robust Features-based image mosaic method for large-scale microscopic hyperspectral pathological imaging

Abstract

Microscopic hyperspectral imaging technology has been widely used in pathological analysis as it can obtain both spatial and spectral information of samples. However, most hyperspectral imaging systems can only capture images in a single field of view. Therefore, an image mosaic is one of the most important steps in a large-scale microscopic hyperspectral imaging system. This paper proposes a microscopic hyperspectral image (HSI) mosaic method based on Speeded-Up Robust Features (SURF) and linear synthesis to achieve large-scale HSIs. In contrast to other SURF-based image mosaic methods, the proposed method leverages both image content and coordinate information to improve the accuracy and stability of the image mosaic. In addition, multiple bands of HSIs with different texture information and grayscale are applied in image matching to take full advantage of spatial redundancy. Simultaneously, a blank microscopic HSI screening method is introduced in this paper to pick out a clearer blank image for better preprocessing, i.e. removing interference in the optical path and the interference of dust on slides. Finally, the preprocessed images are synthesized by linear-based synthesis methods due to their simple synthesis structure and better universality. Additionally, a file format, i.e. hyperslide, is defined for large-scale HSIs and can be browsed with HyperViewer software. Experimental results show that the proposed microscopic HSI mosaic method can obtain high-quality large-scale microscopic HSIs of tissue sections.