Adaptive feature selection and constrained weak-membrane optimization for boundary detection

In this paper, we present an algorithm that locally selects the image features with good discriminatory capabilities and uses a multilayer constrained weak-membrane for boundary detection. We use a general non-stationary image model where image features are allowed to vary slowly within regions and boundaries are defined by abrupt changes in these features. Image features usually have varying discriminatory properties in different regions and the purpose of feature selection is to convey the most discriminatory information for edge detection. The variances of image features are calculated on large and small windows and the difference of these variances are used as a measure of the discriminatory weight of the feature. A layer of constrained weak-membrane is used for each feature and these layers collaborate at every iteration to produce the global discontinuities. Local line configurations are used to constrain the weak-membranes to break only along certain configurations that conform to expected shapes of the boundaries