A method based on multispectral imaging technique for White Blood Cell segmentation

Abstract

Aiming at the problem of image analysis of White Blood Cells in bone marrow microscopic images, multispectral imaging techniques with spectral calibration method was proposed to acquire device-independent images, which is almost impossible in conventional color imaging method. For image segmentation, Support Vector Machine (SVM) was applied directly to the spectrum of each pixel, and using sequential minimal optimization (SMO) algorithm for feature selection to reduce the time of training SVM classifier. Mass of experiments showed that the method is robust, effective and insensitive to smear staining and illumination condition.