سال انتشار: ۱۳۹۰

محل انتشار: هفتمین کنفرانس ماشین بینایی و پردازش تصویر

تعداد صفحات: ۶

نویسنده(ها):

Shaadi Shidfar – Neurosciences and Medical Image Computing GroupUniversity of KentCanterbury, UK
Alan Colchester – Neurosciences and Medical Image Computing GroupUniversity of KentCanterbury, UK

چکیده:

There have been notable advances in high levelsegmentation techniques in different medical applications, butthere has been less progress in general-purpose low levelsegmentation. Data-driven segmentation is needed forconstructing object-based image descriptions, which allowsdetection and categorisation of novel objects and for high-levelmodel matching. A promising approach for the first stages ofsuch a system is based on separatrices. However, significanttopological problems can arise when the approach is used withreal, discretely sampled images, in particular relating to contactsbetween separatrices. We analyze these contacts systematically ina selection of 2-D images and identify the occurrence ofseparatrix crossings. We show that these generate invalid slopedistricts and propose an algorithm for their correction