In so far as geometric shape is concerned a curvilinear band is very similar to a road segment. Roads may be regarded as
a particular type of curvilinear bands. Many methods have been developed for extracting such targets, some of which
are based on edge information. It is very difficult to extract roads and curvilinear bands with no clear and continuous
boundaries using existing algorithms. To mitigate this problem a new strategy is presented in this paper. Edge lines are
detected by means of an improved Burns' method and Dynamic Programming. Target segments are then formed by
matching these edge lines. Connecting the obtained segments produces the final results. Experiments prove the
efficiency of the proposed method while being applied to extract complicated roads and curvilinear bands in different
kinds of images.