Objective: At present, there are three types of\r\nfusion operators for wavelet image fusion: pixel,\r\narea and region. Pixel operators can quickly get\r\na fused image, but the image can be blurred.\r\nArea operators consider the neighboring gray\r\nvalue, which can reduce the edge''s sensitivity\r\nand the fused image has better vision\r\ncharacteristics. Region operators should operate\r\nimage segment to source image. Though it can\r\nobtain an image with the best vision\r\ncharacteristics, the speed is very slow, so it is\r\nnot suitable for medical image fusion, as\r\nmedical image fusion should consider the time\r\ncost. So the aim of the study is to find a proper\r\nfusion operator for medical image fusion that\r\nconsiders not only the time cost but also the\r\nquality of the fused image.\r\nMethods: By analysis of the three fusion\r\noperators, a new fusion algorithm based on\r\nwavelet transform was proposed: low frequency\r\nsub-band based on window neighborhood\r\nentropy larger and high frequency sub-band\r\nbased on window standard deviation larger.\r\nWhat''s more, in order to ensure the consistency\r\nof the fusion image data, it was tested using the\r\nconsistency validation method.\r\nResults: The algorithm was validated by the\r\nCT/PET images. For each image, bior3.9 was\r\nused to perform 4 levels of decomposition, and\r\nsubjective and objective evaluation methods\r\nwere combined to evaluate the fused image. It\r\nwas easy to see the level 4 decomposition had\r\nthe best performance.\r\nConclusion: Experimental results show this\r\nalgorithm can make full use of the correlation\r\nbetween adjacent pixels, and can extract useful\r\ndetail information from the source images: the\r\noutline of the image from CT and the high light\r\nspot from PET.
Loading....