A statistical methodology for the global and local analysis of star tracker image content is presented that is based on the A-Contrario framework. A level set analysis using this methodology effectively weights signals with a confidence interval based on the information content. Globally this analysis can represent the non-planar noise floor associated with the sky background. Locally, this analysis can automatically define the annulus that represents the partial pixels associated with the boundary between signal and noise. The performance of centroiding with information theoretic weighting is evaluated compared to traditional thresholding methods for simulated and real star tracker images.