Research On Pixel-level Image Fusion And Its Application

There are different characters of images which have been achieved by different modalities. Combining relevant information in two images of CT and MRI into a single highly informative image can give much more effective information in clinical diagnosis and treat. The pixel-level image fusion can provide more abundant, accurate and reliable detail information than the fusion at feature-level or decision-level. An intensive research on the methods of CT and MRI grey image fusion and clinical application in pixel level has been made in this dissertation.Different image formats influence the qualities of image fusion. Selecting common image formats such as BMP, JPG and PNG, fused images and objective evaluation criteria are achieved using wavelet transform firstly, and than the comparison experiments are done based on nonsubsampled contourlet transform (NSCT). Simulation results show that PNG format is optimization in medical image fusion.The author introduces the theory of the wavelet transform and multi-resolution analysis, and elaborates the influence factors of medical image fusion based on wavelet transform in this paper. The methods how to select the best wavelet function and decomposed level are given, and then the qualities of the algorithms based on wavelet transform have been analyzed and compared. Comparing the traditional wavelet transform algorithm adopting the"average"rule to low-frequency and the"coefficient absolute value"to high-frequency, an efficient fusion method is proposed which the"region energy"fusion rule is adapted to the low-frequency coefficients and the"region variance"fusion rule to the high-frequency coefficients. Then the further research has been made and the better algorithm which the"region energy"and the"coefficient absolute value"fusion rules are used as the low-frequency and the high-frequency coefficients individually has been introduced. Experimental results indicate that the proposed algorithm of image fusion is effective.The characters of the nonsubsampled contourlet transform(NSCT) are analyzed and the effects of image fusion using different filters,decomposing levels and fusion rules are deeply discussed. After a comparative analysis is carried out of the different classical fusion methods, an efficient fusion method comparing the traditional nonsubsampled contourlet transform algorithm is proposed which the"region energy"fusion rule is adapted to the low-frequency subband coefficients and the combining"region variance"with the"coefficient absolute value"fusion rules to the high-frequency directional subband coefficients. Experimental results show that choosing an appropriate combined filter and fusion rules can get a satisfying fusion result,with less decomposing levels and directional banks,which can reduce the fusion algorithms complexity effectively.The fusion images of CT and MRI play an important role in clinical application. But it is big problem to verify the same level of the scanning images because of the different factors of CT and MRI. So it is very difficulty to be used in image matching and fusion. A new method of real-time image fusion of double machines based on tris-axes landmarks is proposed. As condition and powerful tools, two simulation systems of CT and MRI scanning workstation are realized. The designing of radiotherapy treatment plan system is the important application field and can embody the true value of image fusion. So the radiotherapy treatment plan system has been fulfilled based on 2D image fusion.There are 29 images, 22 tables and 187 references in this thesis.