Spectral feature extraction of blood cells based on hyperspectral data

The aim of this paper is to investigate how to extract spectral features of five kinds of blood cells before they are classified and counted. Here the hyperspectral images of blood cells are taken from leukemic patients and healthy persons. Different from traditional methods which are usually based on morphology, the method of feature extraction based on hyperspectral imaging technique is mainly from spectral pattern traits and similarity measures. The spectral pattern traits in this paper includes characteristics of troughs or crests in spectral patterns such as spectral absorption index, location, intensity, symmetry. Similarity measures are to measure two pattern´s relationship using spectral angle mapping, correlation coefficient and covariance. Altogether these features contain more than thirty characteristics, which would be of great use to segment and classify blood cells in later work.