Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 45435 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Fatemizadeh, Emad
- Abstract:
- Medical imaging is of interest because of information that will provide for doctors and registration is inevitable when we need to compare two or more images, taken from a subject at different times or with different sensors or when comparing two or more subjects together. Registration methods can be categorized in two major groups; methods based on feature and methods based on intensity. Methods in first group have three steps in common: feature extraction, finding matches and transform estimation. In second group it’s important to define a similarity measure and find the transform that minimizes this measure.
Manifold learning algorithms are mostly used as a dimensionality reduction tool and they’re powerful in recovering the underlying structure of high dimension data in low dimension space. Because of this property, these methods are used in image processing to find structures in images.
In this dissertation image registration and manifold learning is introduced then a method for using manifold learning in mono-modal image registration is discussed. This method acts as a preprocessing of images based on landmark extraction and leads to more accurate registration at the end of process. Proposed method is tested for noisy images with different levels of Gaussian noise and also in presence of bias field. Test results are satisfactory and from 5 to 10 percent increase in accuracy is seen.
A novel approach for manifold learning in multi-modal image registration is also proposed in next chapter. Base on intrinsic property of manifold learning method in uncovering the underlying structure, similarity methods can be defined considering the fact that images from different modalities have same structures. These methods can be categorized in three main groups; first group consider differences in Laplacian matrix of both images, in second group manifold of first image is learnt subject to minimizing distance with second image and in third group manifold of both images are learnt simultaneously minimizing the distance of learnt manifolds. Proposed similarity measures are tested in solid registration of MRI and PET images and results show these registrations are exact and correct according to previous measures such as mutual information - Keywords:
- Image Registration ; Manifold-Based Learning ; Similarity Measure ; Medical Images ; Isomap Algorithm ; Laplacian Eigenmaps
Digital Object List
-
محتواي کتاب
- view
Bookmark