Pyramid Structure and Edge Membership Based Blurred Image Reconstruction Algorithm
Date Issued
2012
Date
2012
Author(s)
Chang, Wei-De
Abstract
With the progress of the technology, the requirement of the image quality is getting higher and higher, therefore, the blurred image processing is getting more important. In this thesis, the blurred images from microscope system are used. Similar to ordinary blurred image, the blurred images from microscope system are also created by a point spread function (PSF).
In the microscope system, a true PSF from designed lens can almost be obtained. According to the known PSF, the image restoration procedure become to a non-blind deconvolution problem. However, even if the blurring kernel is known, we cannot easily restore an image perfectly.
In the non-blind image reconstruction, Richardson-Lucy (RL) algorithm, classified to regularization methods is a popular approach for image reconstruction, which uses more iteration times for the sharpness of edge regions in exchange. However, it is a tradeoff that if more iteration times we use, the more undesired ringing artifacts will be produced. In this thesis, to reduce the ringing artifacts, an image deblurring method of pyramid based RL algorithm is proposed. The proposed methods reduce the ringing artifacts and contain the sharpness of edges. Moreover, the computational complexity is also significantly reduced.
Moreover, the Wiener filter is another popular approach in the non-blind image reconstruction. However, due to the uncertainty of noise, the performance of the Wiener filter is restricted. In this thesis, we proposed an edge-membership based image reconstruction algorithm, which focus on the noise uncertainty problem produced from derived Wiener filter. Furthermore, the membership concept is also applied. The experimental results show that our approach has good performance of restored image.
Subjects
blurred image reconstruction
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
index.html
Size
23.27 KB
Format
HTML
Checksum
(MD5):bed9929d9ae0628d25dfbc99b64e7c32