Segmentation of brain MR images using local multi-resolution histograms

Abstract

Histogram based technology has been wildly used in MR image segmentation. However a single histogram suffers from the inability to encode spatial image variation. In this paper, we propose a simple yet novel classification method based on local multi-resolution histograms that utilize the difference among multiple histograms of consecutive image resolutions. We have designed such histogram based attribute vectors which have many desired properties including that they are easy to compute, translation and rotation invariant, and sufficient to encode local spatial information. Comparisons between different methods for expec-tation-maximization-like (EM-like) procedures and hidden markov random field method for simultaneous parameter estimation and partial volume estimation are reported. Experimental results show that our method allows for more accurate and robust classification for brain tissue than these algorithms.