Détection de contours dans les images à l'aide de lignes géodésiques et lignes de champs
Yang, Fang · HAL (Le Centre pour la Communication Scientifique Directe) · 2017
This thesis aims at providing methods for finding contours in images and is composed by two parts: geodesie distance and curves computations using heat diffusions and a boundary-tracking method named point flow. ln the first part of this thesis. we apply different geodesie metrics for heat equation to approximate the geodesie distance. Additionally, we propose two algorithms for extracting the geodesie lines automatically. At last. we come up with an approach for the segmentation of tubular structure by adding an additional dimension to the 2D heat equation. The experimental results testify the robustness and effectiveness of our heat-based method. For the second part of this thesis, we are interested in one of the most fundamental problems in computer vision: edge detection. We propose a model called point flow which can simulate the process of tracking abject boundaries automatically. On the way of finding contours, we do not only consider the gray-level information, but also combine different features to detect the edges and contours. Moreover, this model can be a Iso used to inter the illusory contours. We test our edge detection method on a weil known dataset and the result is comparable to the other classical edge detectors.