Color image segmentation based on the normal distribution and the dynamic thresholding

A new color image segmentation method is proposed in this paper. The proposed method is based on the human perception that in general human has attention on 3 or 4 major color objects in the image at first. Therefore, to determine the objects, three intensity distributions are constructed by sampling them randomly and sufficiently from three R, G, and B channel images. And three means are computed from three intensity distributions. Next, these steps are repeated many times to obtain three mean distribution sets. Each of these distributions comes to show normal shape based on the central limit theorem. To segment objects, each of the normal distribution is divided into 4 sections according to the standard deviation (section 1 below -sigma, section 2 between -sigma and mu, section 3 between mu and sigma, and section 4 over sigma). Then sections with similar representative values are merged based on the threshold. This threshold is not chosen as constant but varies based on the difference of representative values of each section to reflect various characteristics for various images. Above merging process is iterated to reduce fine textures such as speckles remained even after the merging. Finally, segmented results of each channel images are combined to obtain a final segmentation result. The performance of the proposed method is evaluated through experiments over some images.