1) The document presents a method for classifying burn images using a one-class support vector machine (SVM). Burn images are first acquired and preprocessed before feature extraction.
2) An adaptive one-class SVM classifier is used to identify the degree of burn based on whether the images fall inside or outside the SVM boundary. This allows burn images of uncertain classification to be raised to a higher severity level based on expert feedback.
3) The experimental results on a dataset from Cho Ray Hospital show that the adaptive one-class SVM approach is effective for classifying burn images into different severity levels. Future work will focus on improving classification accuracy and expanding the approach to more burn severity levels.
OpenChain - The Ramifications of ISO/IEC 5230 and ISO/IEC 18974 for Legal Pro...
Burn image classification using support vector machine
1. International Conference on Context-Aware Systems and
Applications ICCASA 2015
BURN IMAGE CLASSIFICATION USING ONE-CLASS
SUPPORT VECTOR MACHINE
Author: Hai Tran
Triet Le
Thai Le
Thuy Nguyen
7. Burn Images
4 degrees of burn:
12/7/2015 ICCASA 2012 7
* The Journal of the Ameriacan Medical Association (Nov. 2014),
http://jama.jamanetwork.com/
9. Burn Image Classification process
12/7/2015 ICCASA 2015 9
Step 1: Burn Image Acquistion
Step 2: Drop to standarize the size to segmentation
Step 2: Feature extraction
Step 4: Using SVM classifier to identify the degree of
burn
12. Adaptive SVM classifier for burn images
{x|(wT.x)+b= -1} {x|(wT.x)+b= +1}
{x|(wT.x)+b=0}
+
++
+
+
+
+
+
+
-
-
-
-
-
-
-
-
-
{x|(wT.x)+b= -1} {x|(wT.x)+b= +1}
{x|(wT.x)+b=0}
N/A
o
o o
o
o
In case difficult to identify belong
class II or class III or N/A. Based on
expert suggestion, the computer
system should raise the high level.
Traditional Adaptive
+
++
+ +
+
+
-
-
-
-
-
-
15. Experimental Results on Cho Ray supplying Dataset
http://fit.hcmup.edu.vn/medical_image_project/
16. COCLUSION AND FUTURE WORK
1. Conclusion
- Researching and selecting the suitable SVM Classifier for
burn image classification.
- Adjust OAA strategy of SVM for burn image
classification.
2. Future Work
- Improving the accuracy of SVM classification model for
burn images.
- Increase the number of classes.
- Apply on the bigger data.
ICCASA 2015
17. REFERENCES
1. Janet, M., Torpy, M.D: Burn Injuries, the Journal of the American Medical Association
(JAMA), Vol 302, No. 16, doi:10.1001/jama.302.16.1828 (2009)
2. Michał S.:Introduction to Medical Imaging, Biomedical Engineering, IFE, 2013
3. Survana, M., Sivakumar Niranjan, U. C.: Classification methods of skin burn images.
IJCSIT (2013)
4. Acha, B., Serrano, C., and Laura M.R: Segmentation and classification of burn images by color and texture information.
Journal of biomedical optics 10.3 (2005)
5. Guerbai, Y., Youcef C., and Bilal H.: The effective use of the one-class SVM classifier for
Handwritten signature verification based on writer-independent parameters." Pattern
Recognition (2014)
6. Chebira, A., Kovačević, J.. Multiresolution techniques for the classification of bioimage and biometric datasets. In Optical
Engineering+ Applications (pp. 67010G-67010G). International Society for Optics and Photonics.(2007)
7. Tam, T. D., and Binh, N. T.: Efficient Pancreas Segmentation in Computed Tomography
Based on Region-Growing. Nature of Computation and Communication. Springer
International Publishing, 332-340 (2014)
8. Bao, P. T.: Fast multi-face detection using facial component based validation by fuzzy logic. Proceedings of the
International conference on Image Processing and Computer Vision (IPCV’06), Las Vergas, Nevada, USA (2006)
9. Thai, L. H., Hai, T. S., Thuy, N. T.: Image Classification using Support Vector Machine and
Artificial Neural Network. I.J. Information Technology and Computer Science, Vol. 5, pp.
32-38, DOI: 10.5815/ijitcs.2012.05.05 (2012)
10. Van, H. T., Tat, P. Q., Le, T. H. Palmprint verification using GridPCA for Gabor features. In Proceedings of the Second
Symposium on Information and Communication Technology (pp. 217-225). ACM (2011).
12/7/2015 ICCASA 2015 17