Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Fundus Image Segmentation Based on SVM and Template Matching

Author WuXun
Tutor DongYi
School Huazhong University of Science and Technology
Course Communication and Information System
Keywords Vessel segmentation Template matching Direction operator Support Vector Machine Relaxation factor Morphological operations
CLC TP391.41
Type Master's thesis
Year 2011
Downloads 19
Quotes 0
Download Dissertation

The fundus image segmentation has been the focus of the hot issues of the medical field , a lot of research on retinal vascular segmentation , but most of the algorithms are only good performance in high - contrast crude vessel segmentation for low contrast of fine blood vessels the performance is unsatisfactory, found pathological causes for this type of segmentation of blood vessels are often more challenging , however , is often more persuasive . In this context , we proceed from the theory and application of some methods , using step-by-step segmentation method to extract targeted vascular . Work and innovation of this paper is summarized as follows : (1 ) for the noise characteristics of the image of an improved median filtering algorithm to eliminate most of the noise at the same time try to keep blood vessels of the characteristics of the original combined linear tubular characteristics of the vascular , the use of an adaptive local thresholding algorithm , extract the broad contours of the vascular , to maximize retention vascular prototype . (2) using two methods, the step-by-step on the image contrast enhancement, the balanced background gradation differences , histogram standardized operating as much as possible to increase the contrast between the small blood vessels and the background , reducing the localized differences in the background , the use of different methods to extract image features as support vector machine training samples . (3) to mathematical morphology operation based , on the classified image post-processing , including noise filtered off , to empty the filling of the connection with the slit , and the obtained result of the processing is compared with the other stars Algorithm . As used herein, the experimental data provided from Hoover STARE fundus database , the results show that the methods in this article can be effectively extracted from the fundus image in the vascular structure .

Related Dissertations
More Dissertations