A Medical Image Fusion Algorithm Based on Nonsubsampled Contourlet Transform and Feature Matching

Abstract

By analysis the regional characteristics of Low-frequency subbands and high-frequency subbands in nonsubsampled Contourlet transform (NSCT) for medical images, we proposed an NSCT-based image fusion algorithm for medical images.For each coefficient of the low-frequency subbands, the regional correlation was considered and the fusion strategy based on regional clarity matching was used; for each coefficient of the high-frequency subbands, the directional characteristics of the subbands were considered and the fusion strategy based on local energy of high-frequency regions was used; after that, the high-frequency fusion coefficient was determined.The proposed algorithm was verified through a simulation experiment on CT, PET and MRI images.The fusion effect was assessed using subjective and objective evaluation indicators.The experiment showed that the proposed algorithm achieved a better visual quality and quantitative indicators for medical images. 1.IntroductionFusion of medical images is to integrate the images collected by the same or different techniques on the same target object into one image by extracting and processing the useful information from each image using the image fusion algorithms.Several medical imaging techniques are now in used based on the combination of computer and medical sciences, such as computed tomography (CT), magnetic resonance imaging (MRI) and PET scan.Through the fusion of medical images, several images collected by different techniques are synthesized into a single target image, where the information on bones, soft tissue information as well as their physiological functions is well preserved.Wavelet transform (WT) has found extensive applications in the fusion of medical images [1][2].Ranchin T. and Wald L.[3] applied the discrete WT technique to multi-source image fusion in 1993.WT technique has the advantages of excellent time-frequency analysis performance, directional anisotropy and relative independence of different scales.Therefore, WT technique can achieve a better fusion effect than Laplacian Pyramid transform [4].However, WT involves complex procedures and takes a lot time and storage space.Pennec and Mallat[5] proposed Bandelet transform, and Candes[6] proposed curvelet transform.In 2002, Contourlet transform was presented by M.N.DO and Vetterli M.[7], and wavelet-based Contourlet transform (WBCT) was proposed by Eslami R. and Radha H.[8] in 2004.Nonsubsampled Contourlet transform (NSCT) was put forward by Cunha and Zhou J. P. et al. [9] in 2006.Multi-scale geometric analysis is to achieve an optimal approximation of high-dimensional functions based on its multi-resolution feature and anisotropy.It is more applicable to the representation of sparse 2D image signals than WT.Utilizing the multi-scale, multi-directional, anisotropy and translation invariance features of NSCT, we proposed a medical image fusion method which integrates NSCT and regional feature matching algorithm.NSCT was first done to the source images to obtain low-frequency and high-