Please use this identifier to cite or link to this item:
|Title:||Medical image segmentation : a watershed-based algorithm||Authors:||Bou Habib, Marc||Advisors:||Abche, Antoine||Subjects:||Image processing--Digital techniques
Diagnostic imaging--Digital techniques
Image segmentation is one of the fundamental steps in digital image processing. It is also the most difficult task. The primary objective of this thesis is to develop a robust image segmentation method. It is based on the watershed technique. The algorithm is developed to overcome the drawback of over segmentation of the watershed technique. The thesis proposes a sequence of steps for the segmentation process. Having acquired the original image; the Eigen image, corresponding to the maximum eigenvalue, is generated. Then, the latter image is processed using a gradient technique (a combination of Sobel and Laplacian), and the resulted image is thresholded. The latter image is fed to the watershed algorithm. Since the watershed image yields and over detected Region of Interest (ROI), the pixels that have intensities within a one σ from the mean are kept and the expansion procedure is performed to expand the contours. The expansion process is based on standard deviation in conjunction with an optimized discriminant function. At this point, a manual threshold can be selected by trial and error for the discriminant function. This requires the intervention of the doctors or medical practitioners to check if the selected threshold yields the best segmentation. The procedure is tedious and not recommended. Thus the segmentation is finalized by refining the contour of the image that is resulted from the expansion procedure. This step is also based on the incorporation of the discriminant function(s) associated with the layers of pixels surrounding the boundaries and a threshold that is computed from the evaluated function(s). The proposed approach is evaluated visually and quantitatively on two MRI images of the brain .The results show that the algorithm yields a high correct detection of the manually extracted ROI with a low percentage of over detected and under detected pixels. Therefore, the algorithm is proven to be robust and reliable in segmenting the ROI of an Image.
Includes bibliographical references (p.70).
Supervised by Dr. Antoine Abche.
|URI:||https://scholarhub.balamand.edu.lb/handle/uob/3346||Rights:||This object is protected by copyright, and is made available here for research and educational purposes. Permission to reuse, publish, or reproduce the object beyond the personal and educational use exceptions must be obtained from the copyright holder||Ezproxy URL:||Link to full text||Type:||Project|
|Appears in Collections:||UOB Theses and Projects|
Show full item record
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.