Multiscale Classification Likelihood Estimation of Weak Boundary through WDHMT Model

This paper presents a novel multiscale classification likelihood (MCL) estimation method using hierarchical wavelet-domain hidden Markov tree(WDHMT) model. The key idea is that with inter-scale communication and intra-scale interaction of the WDHMT model, we can capture hierarchical classification information of the pixels at the vicinity of the weak boundary. Our framework consists of the following steps. Each frame extracted from the image sequence is transformed through discrete wavelet transform to obtain a compressive representation of the original one. Then the wavelet coefficients at each scale are represented by a tree-structured probabilistic graph, namely, hidden Markov tree. After the model parameters are learned through up-down iterated expectation maximization (EM) algorithm, we deduced the classification likelihood information at each scale. Finally, we tested the performance of this algorithm by using a sequence of tobacco leaf images, in which the objects are shaded and the boundaries are relatively weak, with encouraging results.