A novel automated approach of multi-modality retinal image registration and fusion

Biomedical image registration and fusion are usually scene dependent, and require intensive computational effort. A novel automated approach of feature-based control point detection and area-based registration and fusion of retinal images has been successfully designed and developed. The new algorithm, which is reliable and time-efficient, has an automatic adaptation from frame to frame with few tunable threshold parameters. The reference and the to-be-registered images are from two different modalities, i.e. angiogram grayscale images and fundus color images. The relative study of retinal images enhances the information on the fundus image by superimposing information contained in the angiogram image. Through the thesis research, two new contributions have been made to the biomedical image registration and fusion area. The first contribution is the automatic control point detection at the global direction change pixels using adaptive exploratory algorithm. Shape similarity criteria are employed to match the control points. The second contribution is the heuristic optimization algorithm that maximizes Mutual-Pixel-Count (MPC) objective function. The initially selected control points are adjusted during the optimization at the sub-pixel level. A global maxima equivalent result is achieved by calculating MPC local maxima with an efficient computation cost. The iteration stops either when MPC reaches the maximum value, or when the maximum allowable loop count is reached. To our knowledge, it is the first time that the MPC concept has been introduced into biomedical image fusion area as the measurement criteria for fusion accuracy. The fusion image is generated based on the current control point coordinates when the iteration stops. The comparative study of the presented automatic registration and fusion scheme against Centerline Control Point Detection Algorithm, Genetic Algorithm, RMSE objective function, and other existing data fusion approaches has shown the advantage of the new approach in terms of accuracy, efficiency, and novelty.

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Sciences

Grantor

Computer Science

Year dc:date.available

2008

Author dc:creator

• Cao, Hua

• Biomedical Image Fusion
• Biomedical Imaging
• Adaptive Exploratory Algorithm
• Control Point Detection
• Biomedical Image Registration
• Mutual-Pixel-Count
• Heuristic Optimization Algorithm

Statement dc:rights

• unrestricted
• Release the entire work immediately for access worldwide.

Identifier

etd-02262008-155553
https://repository.lsu.edu/gradschool_dissertations/735

OAI identifier oai:identifier

oai:repository.lsu.edu:gradschool_dissertations-1734