Adaptive Bezier Curve-based Membership Function formulation scheme for Interpretable Edge Detection

Abstract

The quest for the design of interpretable models is expanding as there is a far-reaching reliance on Black box models, owing to their lesser interpretability in predictions. This paper presents an interpretable fuzzy logic model for edge detection based on a novel generalized domain independent parametric Adaptive Bezier Curve-based Membership Function (ABCMF) constructed for image fuzzification. To bring out a robust fuzzy framework using the developed novel membership function, the fuzzified image is convolved with the oversimplified fuzzy-based edge detector to determine the direction of intensity changes. Finally, using the λ− cut technique, the edge detected image is transformed back to crisp form with adaptively varying λ. The efficacy of the proposal is exhaustively tested on BSDS500, BIPED, and MDBD, and the attained simulation results are compared with traditional and current methods. From the evaluated metrics, it can be inferred that the proposed method offers a consistent accuracy greater than 91% in comparison with its counterparts. Also, when analyzing in terms of interpretability, fairness, F1−score, and computational efficiency, the approach offered an increment of 6% when compared with recent models.