Image Fusion Based on WRV and NHM Rules for Medical Digital Image in Contourlet Domain

With the rapid development of medical information management, there are an increasing number of medical digital images, and clinicians need to integrate comprehensive multi-source images to make an accurate diagnosis of a patient. To improve the ability to understand the fused information of multi-source medical images, a technique for image fusion based on weighted regional variance (WRV) and neighborhood homogeneous measurement (NHM) rules was proposed in this article. First, the source images were decomposed in multi-scale and multi-direction to obtain sub-band coefficients by conducting contourlet transform. Then, by analyzing the coefficients, the appropriate fusion rules can be selected. The WRV rule was performed to fuse low frequency subband coefficients as the characteristics of low frequency sub-band mainly reflect the contour of an image. Additionally, the NHM rule was adopted to integrate high frequency sub-band coefficients which contain a lot of useful edge details of source images. Finally, an experiment based on computed tomography (CT) and magnetic resonance imaging (MRI) medical images was conducted. We also used evaluating indexes to verify the performance of the proposed algorithm. The simulated results show that the proposed method can extract features from source images and inject it into the ultimate object image, and evaluations can also prove that the method improves the blurring problem on edges and contours and promotes the performance of the algorithm. Subject Categories and Descriptors H.2.8 [Database Applications]: Image database; I.5.3 [Pattern Recognition]: Algorithms Juan WANG1, Siyu LAI2, Borjer T.H3 1Computer School, China West Normal University Nanchong, 637002, China 2Imaging Department, North Sichuan Medical College Nanchong, 637000, China 3Information Institute for Geographic Engineering in ABB Communication Corporation Patras, 138682, Greece [email protected] General Terms: Medical Imaging, Algorithm, Frequency Domain, Evaluation Indexes