Robust edge detection

摘要

In this paper, we have described a new robust edge detection algorithm which performs equally well under a wide variety of noisy situations and a broad range of edges. The algorithm is executed in three phases. In phase 1, the step and linear edges are detected from the noise corrupted image. The corrupting noise is assumed to be additive with two components: (1) A symmetric, continuous, exponential type noise component such as Gaussian, Laplacian or uniform noise and (2) an impulsive noise component. While well-known edge operators use smoothing before edge detection, phase 1 of the new operator uses a statistical classification technique to detect the step and linear edges. The statistical classification technique is based on two salient characteristics of natural edges: (1) Near and around the step and linear edges, the pixels, when classified into two nearly equal groups, display markedly different average intensity values, (2) the members of each group show strong spatial correlation. The advantage with the new technique is that the weak edges are not blurred, and the presence of impulse noise is easily accounted for. In phase 2, all the thin-line edges, i.e. edges which are lines less than two pixels wide, are detected by a supplementary technique since these edges cannot be detected simultaneously with the other step and linear edges. In order to do so, the problem of thin-line edge detection is posed as a series of outlier detection problem. The solution is provided in the framework of Dixon's r-test. In phase 3, the spurious edge elements are suppressed, and the isolated missing edge elements are interpolated using a number of hypothesized edge-segments. Finally, some experimental results are provided to illustrate the success of the algorithm.