Active Contour Image Segmentation in Fisher Discriminant Spaces

Abstract

In this paper, we introduce an algorithm that is able to segment objects in natural images by using active contours. Active contours are used to regularize the segmentations. Our approach utilizes multiple feature spaces to capture as much information as possible, followed by projecting the multiple dimensional features space onto a single dimension to enable improved active contour evolution. We apply the Fisher Linear Discriminant Analysis (FLDA) to optimally calculate the projection vector while providing prior knowledge on number of clusters that are present on the image. Preliminary experiments confirm that the proposed algorithm is able to segment objects in natural images while optimizing contour smoothness and noises.