Modification in the Contour Relaxation Model


It was shown in section 2 (Figure 1) how application of the contour relaxation resulted in partitioning of the sky region into a number of smaller regions (resulting in false contours). We propose a modification in the contour relaxation model in order to avoid such an image partitioning. The modification involves introduction of a bias in the likelihood function which suppresses the false contours formation process by adjusting the likelihood function ratio according to the change in the region size. The bias is modeled by the following function:

f(s)=exp(-w(<>sjk/(sjo+<>sjk)))

where sjo is the jth region size at the zeroth iteration, <>sjk is the change in the jth region at the kth iteration of the relaxation.

Instead of maximizing the ratio r(Y,Q) of the likelihood function p(Y,Q), a new ratio r(Y,Q,s), incorporating the bias f(s) in the model is maximized,

r(Y,Q,s) = r(Y,Q) * f(s)

Where r(Y,Q) denotes the ratio of the likelihood function p(Y,Q).