A Leukocyte Nucleus Segmentation Scheme Based on Fingerprint Smoothing

This paper presents a novel segmentation scheme for leukocyte nucleus based on fingerprint smoothing. An original histogram of a leukocyte image includes many peaks and valleys because of “spike” pulse interference. In this scheme, multi-scale space filter is performed to original histogram and fingerprints are obtained by detecting the location of zero crossings of the derivatives of the signal in scale-space first. Then a proper scale-space value is determined according to the information of the fingerprints, and the leukocyte image is smoothed by Gaussian filters. It not only smoothes out the noise, but also maintains the details. Finally a proper threshold of the smoothed histogram is found to segment the leukocyte nucleus accurately. To obtain more accurate segmentation results, mathematical morphology post-processings are presented subsequently. Experimental results indicate that the proposed scheme is effective and robust.