A New Image Segmentation Technique Based on Non-Parametric Mixture Model
Zhe Liu, Jianguo Xiao · 2010
To solve parameter estimate method's over-reliance on priori assumptions in finite mixture models, the paper proposes image segmentation based on the Laguerre orthogonal polynomial non-parametric mixture model. Firstly, a non-parametric mixture model based on the second Laguerre orthogonal polynomial is designed, and then estimate smoothing parameter of every model with minimum mean-square error (MISE). Secondly, get orthogonal polynomial coefficients and mixing ratio of the models by EM algorithm. The method proposed in the paper overcomes model mismatch without any assumption to the model. Image segmentation experiments shows that the method is more efficient than Gaussian mixture model segmentation, and that it has higher quality than other non-parametric mixture models segmentations do.