Correction of Inhomogeneous MR Images Using Multiscale Retinex

A new method for enhancing the contrast of magnetic resonance images (MRI) by retinex algorithm is proposed. It can correct the blurrings in deep anatomical structures and inhomogeneity of MRI. Multiscale retinex (MSR) employed SSR with different weightings to correct inhomogeneities and enhance the contrast of MR images. The method was assessed by applying it to phantom and animal images acquired on MRI scanner systems. Its performance was also compared with other methods based on two indices: (1) the peak signal-to-noise ratio (PSNR) and (2) the contrast-to-noise ratio (CNR). Two indices, including PSNR and CNR, were used to evaluate the performance of correction of inhomogeneity in MR images. The PSNR/CNR of a phantom and animal images were 11.8648 dB/2.0922 and 11.7580 dB/2.1157, respectively, which were higher or very close to the results of wavelet algorithm. The retinex algorithm successfully corrected a nonuniform grayscale, enhanced contrast, corrected inhomogeneity, and clarified the deep brain structures of MR images captured by surface coils and outperformed histogram equalization, local histogram equalization, and a waveletbased algorithm, and hence may be a valuable method in MR image processing.