|Title:||Automated blood detection for endoscopic images|
|Advisors:||Chu, Henry (ME)|
|Subject:||Hong Kong Polytechnic University -- Dissertations|
Robotics in medicine
|Department:||Faculty of Engineering|
|Pages:||112 pages : color illustrations|
|Abstract:||This dissertation is intended to develop an effective and efficient algorithm for vision-based bleeding detection, applicable in a proper robotic setting to perform automated extra-blood removal as medical assistant tool during an open surgery. To address the primary tasks of automated bleeding detection, drawing accurate bleeding region contours and locating region centroids for tools' approaching, a set of algorithms is developed based on representative images. These images are captured and collected from about 20 endoscopic videos for the diversity of study samples, in turn for the robustness of algorithms. However, in real application scenario, a live video will be continuously taken while images will be extracted from this video as frames for analysis and a feedback mechanism will also be constructed to transform the vision detection result to real-time robotic operation. The images extracted from the same video are more homogeneous in any perspective and can be seen as a subset of the broader research sample group this dissertation attempts to cover. Two approaches are tested to identify which is better in automated detection. The first approach follows normal learning methodology that feeds raw train set of bleeding positive and negative pictures into SVM classifier and make it learn to predict new image for bleeding or not. While the second approach is to identify some interested area on images of both categories beforehand and feeds masked dataset into SVM classifier for learning. The second approach offers superior performance in prediction. In the process of drawing blood region contours and centroids, multiple image processing techniques are utilized. Color constancy is manipulated for frames identification and removal; property of change rate is relied on to balance obvious discontinuity in illuminance; k-means clustering, and image synthesis are used for grippers' removal and image reconstruction. Frequency domain filtering is applied for smoothing followed by dynamic thresholding to generate binary images for delineating contours and locating centroids.|
|Rights:||All rights reserved|
Files in This Item:
|991022270854103411.pdf||For All Users (off-campus access for PolyU Staff & Students only)||5.55 MB||Adobe PDF||View/Open|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: