Color image segmentation is essentially a clustering process in 3D color space

and color images can be considered as s special case of multi-spectral images. It is quite often when the objects cannot be extracted using three features but can be extracted using more than three features. In this paper

every pixel of an image is described using 5 features which are transformed from its own RGB features

then multi-dimensional thresholding(MDT) divide color space by thresholding each component histogram. This approach is equivalent to partitioning the multi-dimensional histogram into rectangular hyper-prisms. But MDT will lead oversegmentation in result. Hence

two-step algorithm is proposed to solve the problem of oversegmentation. The first step is region growing in 3D histogram by connectivity of frequency of neighbor bin in same class. The second step is data clustering based on scale space theory

which models the blurring effect of lateral retinal interconnerctions. The final algorithm that combines different approaches results in further improvement in performance. It is simple and effective to different color image

and has been used to segment the microscope medical image successfully.