THE EFFECT OF CONTRAST ENHANCEMENT METHOD FOR K-MEANS CLUSTERING SEGMENTATION OF WHITE BLOOD CELL CYTOPLASM IMAGE

The appropriate image contrast becomes one of the critical success factors in white blood cell segmentation process because of blood cell composition complexity and background clarity. Segmentation of white blood cells can be divided into two, namely nucleus segmentation and cytoplasm segmentation. Nucleus segmentation cannot be used to obtain the cell cytoplasm, whereas, in cytoplasm segmentation, the nucleus indirectly is also obtained because the nucleus is always in the cytoplasm. This study has successfully compared the effect of three contrast enhancement methods namely top hat and bottom hat transform, linear contrast stretching and fuzzy logic-based image histogram as a
pre-processing stage for K-means clustering segmentation of white blood cell cytoplasm using 15 images of blood sample of RGB, HSV and Lab colour model. The results of the analysis show that the image resulted by pre-processing stage using the top hat and bottom hat transform for an image with RGB colour model yields the highest average sensitivity and accuracy, 80.95% and 99.19%, and it
also has the lowest execution time, 71.06 s. While the highest average value of specificity and the lowest value of deformation, 99.51% and 38.93%, produced by the fuzzy logic-based image histogram method. While for RGB, HSV and lab
variations in linear contrast stretching method, the RGB image resulted best in sensitivity, specificity, accuracy, deformation and execution time. Those are 79.26%; 99.49%; 99.15%; 39.87% and 74.83 s.