Edge Detection of Complex Scenes Based on Bayesian Statistical Inference

Hao Zhang,Jin-hui Cai,Ping-jie Huang,Ze-kui Zhou

DOI: https://doi.org/10.3321/j.issn:1000-565x.2007.09.009

2007-01-01

Abstract:The traditional edge detectors are inefficient in the image detection in complex scenes due to the disturbances of noise and texture. In order to solve this problem, a multi-information fusion edge detection algorithm based on Bayesian statistical inference theory is proposed. This algorithm fuses the output responses of four operators, such as the gradient operator, the Laplacian operator and the ratio of average (ROA) operators at two scales, achieves the optimal discretization for the continuous attributes of feature vector by maximizing the class-attribute mutual information, employs the nonparametric histogram method to estimate the class-conditional probability density functions, and adopts the principle of Bayes Risk Minimization to complete the edge detection of new images. Experimental results show that the proposed algorithm is feasible, with a Bhattacharyya error bound of 0.093 and area under the receiver operating characteristic carve (AUC) of 0.958. The comparison of the detection results obtained respectively from the proposed algorithm and from the classical detectors also shows that the proposed algorithm is robust to the noise and texture in images.