Multiscale Significance Run: Realizing the ‘Most Powerful’ Detection in Noisy Images

Detection is a fundamental problem in many applications. In many cases, knowing the presence of underlying objects is of significant importance. Multiscale methods have been demonstrated to be advantageous in solving this problem. Besides theoretical results that have been achieved, this paper discusses how the ‘most powerful’ detection can be realized, for a set of specifically organized underlying objects. We focus on the design of the detection procedure. Multiscale Significance Run Algorithm—MSRA—serves as a general framework. It is shown that by assigning an hierarchy to the alternatives, one can nearly realize the most powerful detection under certain conditions.