For the music structure segmentation task, one wants to solve two co-existing but sometimes contradicting problems; find global repetitive/homogenous structures and locate accurate local change points. In this paper we propose two algorithms to address these two problems. The algorithms can independently or jointly be plugged into various existing structural segmentation algorithms to improve their results. One algorithm utilizes the Variable Markov Oracle, a suffix automaton for multi-variate time series capable of finding repeating segments, is proposed to obtain a self-similarity matrix which encodes the global repetitive structure of a music piece. The other proposed algorithm is an iterative boundary adjustment algorithm refining boundary locations. The algorithms are evaluated against the Beatles-ISO dataset and achieve comparable performance to stateof-the-art.