Research On Image Fusion And Denoising Algorithm In The Quaternion Wavelet Transform Domain

Image fusion is to merge the advantages information of each image into a newimage to the greatest extent after a certain treatment of the image data on the sametarget that acquired by multi-source channel, thereby improving the utilization of theimage information. Digital image is often affected by interference of externalenvironmental noise in the digitization and transfer process, denoising purpose is notonly effective in removing image noise, and while maintaining the image edges anddetails information.Quaternion wavelet transform is a new kind of multi-scale analysis imageprocessing tool, which is the improvement of the standard wavelets and spread of thecomplex wavelet, and has approximate shift-invariance, so it can overcome theshortcomings of traditional wavelet very well. Moreover quaternion wavelet transformprovides a richer phase information in image analysis. This thesis studies the imagefusion and denoising algorithm based on quaternion wavelet domain. The main work isas follows:1. We overview the development status of image fusion and denoising, and brieflydescribe the image performance evaluation criteria of image fusion and denoising.We introduce the concept of quaternion wavelet, and further focus on theprinciples of quaternion wavelet decomposition and reconstruction.2. For the relationship between the coefficients of multi-focus image and theshortcomings of traditional wavelet in image multiscale analysis, we take fullaccount of the directional relationship of the coefficients and improve the sum-modified-Laplacian, propose multi-focus image fusion method based onquaternion wavelet transform. Experiments show that our method is better than thetraditional methods in subjective vision effect and objective evaluation index.3. Nonsubsampled quaternion contourlet transform is constructed by combining thequaternion wavelet transform and nonsubsampled directional filter banks, whichovercome the shortcomings of the traditional wavelets do not have approximateshift-invariance and multi-direction. Then we propose a new image denoisingalgorithm based on the combination of symmetrical normal inverse Gaussian modelin nonsubsampled quaternion contourlet transform domain and non-local meansfilter in spatial domain. Simulation experiment shows that our method obtains better denoising effect and outperforms other current outstanding algorithms.