الفهرس 
Chapter 1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 215 |  |  |
| | | from | 215 | | |
Abstract
Medical imaging is a very important topic in the area of image processing field. Medical imaging helped doctors to make keyhole surgeries for reaching the interior parts without really opening too much of the body. Some of these medical imaging technologies such as, Computed Tomography (CT) Scanner, Ultrasound and Magnetic Resonance Imaging (MRI) took over X-ray imaging by making the doctors to look at the body’s elusive third dimension.
In this thesis, a digital image interpolation method that is performed in the wavelet domain with a least squares algorithm is presented. This method estimates wavelet coefficients in the high frequency sub-images of the estimated High-Resolution (HR) image from the Low-Resolution (LR) image using a least squares algorithm. An inverse wavelet transform is then performed for the synthesis of the HR image. Experimental results show that the proposed method outperforms other commonly used methods such as the bilinear, bicubic, and traditional least squares methods, objectively and subjectively.
A neural implementation is suggested for maximum entropy image interpolation technique. The performance of the suggested neural image interpolation algorithm is compared to maximum entropy image interpolation technique. The proposed approach is efficient from the mean square error (MSE) point of view and the computation complexity point of view by reducing the number of computations required. Simulation results show that the suggested implementation has little reduced peak signal-to-noise ratio (PSNR) as compared to maximum entropy image interpolation, but with more reduced computational time.
Two suggested approaches are presented in this thesis to obtain high-resolution images from the fusion and then interpolation of Magnetic Resonance (MR) and Computed Tomography (CT) images. MR and CT images are fused with
iv
either the Discrete Wavelet Transform (DWT) or the curvelet transform. After that, wavelet least-squares interpolation step is carried out on the wavelet sub-bands of the fusion result. Also, a neural modeling of the maximum entropy algorithm is used to interpolated the fusion result. Simulation results show the feasibility of the fusion process to obtain images with more details and the efficiency of interpolation to obtain high-resolution images.
Also, an image scale-up (super-resolution) algorithm is introduced based on image fusion principle. Magnetic resonance and computed tomography images are scaled-up using sparse-representation modeling with dictionary learning algorithms. The MR and CT images are fused either by discrete wavelet or curvelet transforms, and the fused result are scaled-up by the same algorithms. Simulation results show that scaling-up the fused CT and MR images, whether they are fused by wavelet or curvelet fusion technique, provides higher PSNR values than scaling-up the CT or MR image separately. Also experimentally we deduce that using the curvelet fusion technique provides better results than using the wavelet, and scaling-up by Michael et al. algorithm gives more better results than bicubic and Yang et al. algorithms in almost all the tested images.
Finally, an image interpolation algorithm based on the contourlet transform is presented in this thesis. The use of the contourlet transform improves the regularity of object boundaries in the generated high-resolution images. An edge-based image interpolation technique that uses wavelet transform with symmetric biorthogonal wavelets is used as the initial estimate for the contourlet interpolation algorithm. An iterative projection process is then used to drive our solution aimed towards an improved high-resolution image. Experimental results show that the proposed algorithm objectively and subjectively outperforms other commonly used algorithms such as the bilinear, and linear wavelet algorithms, and also the image interpolation with geometric representations.