High magnification optical systems such as microscopes usually suffer from limited depth-of-field (DOF) problem. This hinders efficient observation of microscopic objects and prevents vision based approaches from being applied to automatic micromanipulation tasks. A DOF extension method using variable annular pupil was proposed by the authors' previous publication, and a tradeoff between the depth extension range and image quality was observed. In this article, this tradeoff is numerically analyzed by a proposed wavelet-based image quality measure, and a proper DOF scheme to provide the best image for the given object distribution is proposed. The problem is reduced to finding the optimum pupil division parameter. The experimental results conducted for various imaging conditions with some MEMS objects verify that the system DOF can be adjusted according to their position in depth direction to provide the best visibility.