Open AccessLevel Set for Semantic Segmentation with Edge Compensation
Author(s) -
Zhipeng Lei,
Wei Zheng,
Yongxin Miao,
FuZhen Xuan
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1449/1/012041
Subject(s) - initialization , level set (data structures) , segmentation , enhanced data rates for gsm evolution , compensation (psychology) , set (abstract data type) , boundary (topology) , artificial intelligence , computer science , level set method , computer vision , algorithm , pattern recognition (psychology) , intensity (physics) , image segmentation , mathematics , optics , physics , psychology , mathematical analysis , psychoanalysis , programming language
This paper demonstrated that active contour based on points evolution is not suitable for Objects with blurring boundary segmentation. In irregular areas adjacent points have similar motion trends trapped into ‘piles phenomena’. Level set should be preferred in practice. Whereas, classic level set lacks of perception in distance especially narrow and abnormal region. Consequently, we reported an algorithm localized level set that is able to improve accuracy. Meanwhile, in cases lost boundary of bone, we gave a strategy called edge compensation. Depending on shapes of neighborhood slices, defective section is estimated and restored effectively. Our experiments showed that the algorithm localized level set increases segmental quality with precision 99.74%. Additionally, it could not only rectify mistakes brought by incorrect initialization but also have a robust performance to overcome local region with highly inhomogeneous intensity.