Clustering Color Image Segmentation Based on Maximum Entropy
Haifeng Sima, Lanlan Liu · 2012
Maximum entropy is meaningful for representing pixels spatial distribution in the image.This paper proposed a new clustering segmentation approach for color image according to the maximum entropy.Firstly, quantize the HSV color space to equal intervals.The probability distribution of pixels in the quantized space can be seen as a random process.Select a slide interval on the histogram to estimate the classes based on the maximum entropy in the color space.Then observed class number and the initial cluster center.Segmented pixels in to regions by clustering and used spatial filtering to eliminate meaningless regional and holes.The experiment results has shown that this algorithm achieved a good segmentation.