Abstract :
The paper describes a wrapper approach that applies Gauss mixture model into image segmentation, solving the problems of slow segmentation speed, fuzzy contour of the object of interest. By wrapping the feature selection algorithm inside the classifier, i.e, introducing wavelet transform while using EM algorithm to calculate the parameters of GMM image segmentation model, it can get more multi-scale feature information of images, and hence reducing the number of iteration and enhancing the efficiency of image segmentation. The performance of the wrapper-based image segmentation is shown on real-world which proves that the algorithm has better effect of image segmentation.
Keywords :
Gaussian processes; discrete wavelet transforms; expectation-maximisation algorithm; feature extraction; fuzzy set theory; image classification; image segmentation; EM algorithm; Gauss mixture model; feature selection algorithm; fuzzy contour; image classification; image segmentation; wrapper-based DWT-GMM; Discrete wavelet transforms; Educational institutions; Feature extraction; Filters; Image segmentation; Iterative algorithms; Maximum likelihood estimation; Pervasive computing; Signal processing; Signal processing algorithms;