Medical Image Fusion Algorithms Based On Multi-scale Geometric Analysis

With the fast development of technology, many advanced technologies and science achievements are applied to the medical field, and a large number of high-precision medical imaging techniques have been emerged. Because of the different imaging mechanism of the sensors, the physiological information from them has different focus and defects, Multi-modal medical image fusion technology is proposed in order to solve this problem. This technology can provide more comprehensive, more reliable, more visible information for the medical diagnosis and medical measures through extracting and integrating information from different modal medical images. In this thesis, the focus is optimization of the existing algorithm of medical images based on lifting wavelet transform and dual-tree complex wavelet transform,and the main research contents are summarized as follows:1) In the first chapter, the background and significance of medical image fusion are introduced firstly. Secondly, the research status and existing challenges in this field are indicated.At last, it makes a sum up of the innovations in this paper. In the second chapter, it introduces the basic knowledge of medical image fusion firstly, such as image fusion process and so on. Then it studies comparatively of some common medical image fusion and points out the advantages and disadvantages. At last, it proposed a set of fusion quality evaluation system.2) Pointing at the properties of the medical image obtained and medical image specific requirements, a new image fusion algorithm based on lifting wavelet transform and PCNN is proposed. At first, the image is decomposed by lifting wavelet transform into high and low frequency sub-bands. The low frequency regional correlation is determined by regional variance,and the coefficient is determined based on the regional energy. In high-frequency sub-band coefficients fusion rules, the spatial frequency was input to PCNN network as a signal,coefficient is determined based on the ignition characteristics of the region. Finally, the image is generated by lifting wavelet inverse transformation.3) Pointing at the edge characteristic differences of high-frequency sub-band,a new image fusion algorithm based on dual-tree complex wavelet transform is proposed. At first, high frequency sub-bands are classified according to the edge characteristic. The coefficient with strong edge characteristic is determined based on the regional gradient energy, the coefficient with weak edge characteristic is determined based on the regional variance. In low-frequency sub-band, and the coefficient is determined based on the regional energy. Finally, the image is generated by dual-tree complex wavelet inverse transformation.4) In this thesis, lots of experiments of image fusions including gray images and color images are conducted, and the results show that the fused image of this fusion algorithm is more detailed and retains much more edge details with good visual effect.