Abstract
Magnetic resonance (MR) brain images often have low contrast due to different selections of pulse sequence parameters. Histogram equalization method is one of the most popular image enhancement techniques for improving quality of digital images. However, a single histogram equalizer is not able to detect details of contrast existing in the whole image and enhanced results are often distorted. Inspired by machine learning ensemble, a strategy using multiple histogram equalizers on local windows has been applied on enhancing natural images. In this paper, this strategy is investigated further again and applied on MR brain images for contrast enhancement. Each single histogram equalizer on local patch of MR brain image is considered as a single classifier or predictor in machine learning ensemble. A global normalization is used for enhancing the whole image from multiple histogram equalizers. The effect of noise amplification of accelerated MR data acquisition on image enhancement is also studied. The experimental results demonstrate that the proposed method not only enhances contrast among brain tissues (like white matter, grey matter, and cerebrospinal fluid) of MR brain images, but also outperforms other image enhancement methods for improving image quality.