This paper presents a novel method based on MaximumLikelihood Estimation (MLE) to evaluate pixel directivity for depicting image contours of objects as perceived by human eyes. The method is characterized by employing discrete masks with different shapes centered at a target pixel to sample gradient orientations of neighboring pixels for evaluating directivity of the target pixel, and applying MLE to determine one of these discrete sampling masks that best fits the orientation similarity of the target pixel. We show that such a fitting process in effect fulfils the similarity and proximity laws in Gestalt theory, and a salient alignment location can be determined by subjecting the optimal directivity in conjunction with the gradient magnitude of the target pixel to a Bayesian process. Finally, the directivity of salient alignment locations is incorporated with the extension field (Guy and Medioni, 1992) to detect perceptual contours. Experiments tested on complex images and underwater images are provided to justify the superiority of the work over others.

Included in

Engineering Commons