Semantic segmentation by ways of interactive post-processing with active contours model

Abstract

Active contour has been proven to be effective at solving semantic segmentation of images. However, the use cases of such approaches were mostly on trivial problems with narrow utility. For instance, it was used for detecting tumors from MRI scans and finding oil spills from aerial photographs. This thesis considers ways of making active contour work for any kinds of images. Active contour is applied at post-processing step on results from other algorithms.