Real-Time Automatic Linear Feature Detection in Images

Linear feature detection in digital images is an important low-level operation in computer vision that has many applications. In remote sensing tasks, it can be used to extract roads, railroads, and rivers from satellite or low-resolution aerial images, which can be used for the capture or update of data for geographic information and navigation systems. In addition, it is useful in medical imaging for the extraction of blood vessels from an X-ray angiography or the bones in the skull from a CT or MR image. It also can be applied in horticulture for underground plant root detection in minirhizotron images. In this dissertation, a fast and automatic algorithm for linear feature extraction from images is presented. Under the assumption that linear feature is a sequence of contiguous pixels where the image intensity is locally maximal in the direction of the gradient, linear features are extracted as non-overlapping connected line segments consisting of these contiguous pixels. To perform this task, a point process is used to model a network of line segments in images. Specific properties of line segments in an image are described by an intensity energy model. Aligned segments are favored while superposition is penalized. These constraints are enforced by an interaction energy model. Linear features are extracted from the line segments network by minimizing a modified Candy model energy function using a greedy algorithm whose parameters are determined in