Background: Colour image segmentation is fundamental and critical for quantitative\nhistological image analysis. The complexity of the microstructure and the approach\nto make histological images results in variable staining and illumination variations.\nAnd ultra-high resolution of histological images makes it is hard for image segmentation\nmethods to achieve high-quality segmentation results and low computation cost at the\nsame time.\nMethods: Mean Shift clustering approach is employed for histological image\nsegmentation. Colour histological image is transformed from RGB to CIE L*a*b*\ncolour space, and then a* and b* components are extracted as features. To speed up\nMean Shift algorithm, the probability density distribution is estimated in feature space\nin advance and then the Mean Shift scheme is used to separate the feature space into\ndifferent regions by finding the density peaks quickly. And an integral scheme is\nemployed to reduce the computation cost of mean shift vector significantly. Finally\nimage pixels are classified into clusters according to which region their features fall\ninto in feature space.\nResults: Numerical experiments are carried on liver fibrosis histological images.\nExperimental results demonstrate that Mean Shift clustering achieves more\naccurate results than k-means but is computational expensive, and the speed of\nthe improved Mean Shift method is comparable to that of k-means while the\naccuracy of segmentation results is the same as that achieved using standard Mean\nShift method.\nConclusions: An effective and reliable histological image segmentation approach is\nproposed in this paper. It employs improved Mean Shift clustering, which is speed up\nby using probability density distribution estimation and the integral scheme
Loading....