A NOVEL SPATIALLY CONSTRAINED MIXTURE MODEL FOR IMAGE SEGMENTATION

We present a novel spatially constrained mixture model for image segmentation. This model assumes that the prior distribution for each pixel depends on its neighboring pixels', and the degree of dependency is decided by the geometric closeness. The negative log-likelihood function of the proposed method is viewed as energy function, and the parameters of the energy function are estimated by gradient descent algorithm. Evaluation of the developed method is done on synthetic and real world images. Experimental results are compared with those obtained using mixture model-based methods. The proposed approach performs better than other ones in terms of classification accuracy.