Magnetic Resonance Image Fusion using Bilateral Filter and PCA

Three bands of magnetic resonance image—PD, T1 and T2—produce complimentary data about the human body portion being scanned. Medical image fusion combines complementary information of all three bands in the fused image with all relevant information needed by a physician to do medical analysis. This paper uses the edge-preserving bilateral filter and principal component analysis to combine complimentary data from PD, T1 and T2 bands of magnetic resonance images. Bilateral filter is used to decompose each of the three bands of magnetic resonance images into detail and approximation images. Detail image contains the high frequency information whereas approximation image contains the low frequency information. Principal component analysis and maximum absolute value are selected as the activity measure to fuse detail and approximation images. Then, the fused image is obtained by adding these fused detail and approximation images. Fusion metrics like Mutual information, Non-linear correlation information entropy, Structural similarity index measure, Gradient metric, Phase congruency metric and Human perception metric are used in this paper to compare the results of proposed method with image fusion using various pyramids, Wavelet and Curvelet Transforms.