In applications related to object recognition and image understanding it is essential that preprocessing of image data yield meaningful edges and closed contours. An edge extraction and extrapolation approach using both bottom-up and top-down analyses is presented. This approach vectorizes initial edge data and extrapolates the vectors, based on a predefined cost function, in order to link edge data. After the initial clusters of edge vectors are obtained through the linking process, a top-down analysis is initiated to test the validity of the clustering. As a result of this analysis, the extrapolation of edge vectors is modified based on the support of additional edge data. Such a low-level processor provides a logical and systematic method of clustering incomplete edge data, and would be well suited for a cooperative or neural-network-based system which could provide higher level top-down feedback for further analysis.