Reinforcement of Edge Detection by Concurrent Segmentation on Images with Elongated Structures

It is well known that edge detection and segmentation are complementary techniques, one can be deduced and enhanced from the result of the other retrospectively. However, their simultaneous processing can be arranged as well to produce better edges. This paper has proposed a new approach that makes great enhancement on edge detection specifically on noisy images where standard edge detection algorithms based on gradients, fail to track the boundary or branch to a false positive edge. In these cases where the weakened gradients precludes from going ahead, the segmented regions come up on dual sides of already detected edges, and supply needed information to discover the lost clues of edges. The experimental results show that both accuracy and recall values get incremented by 30% utmost on some very noisy images and about 20% on the whole image set.