Bayes factors for edge detection from wavelet product spaces

Interband wavelet correlation provides one approach to defining edges in an image. Interband wavelet products follow long-tailed density distributions, and in such a context thresholding is very difficult. We show how segmentation using a Markov-field spatial dependence model is a more appropriate approach to demarcating edge and non-edge regions. A key part of this work is quantitative assessment of goodness of edge versus nonedge fit. We introduce a formal assessment framework based on Bayes factors. A detailed example is used to illustrate these results.