Texture segmentation using iterative estimate of energy states

A hidden Markov model (HMM) is combined with texture energy transformation to segment textured regions in digital images. Each textured region in an image is modeled as a hidden state of the hidden Markov chain. The probabilistic output at a hidden state is a linear combination of texture energy measurements. The Baum-Welch algorithms is used to estimate the HMM parameters successively. This texture segmentation technique does not depend on prior knowledge of the textural content of the images. Seven masks are used to measure texture energy. To reduce the dependency of a priori choice of the masks used in measuring the energy, two approaches are used: the Karhunen-Loeve transform on texture energy vectors and a double median filter across the estimated edges produced by each mask, and across the masks. This combination of HMM and texture energy works satisfactorily when the test images respond favorably to at least two masks. Some experimental results are presented