Bitangents have been shown to be very useful for planar object recognition. We propose a new algorithm for detecting bitangents in planar curves and compare it to two algorithms used before. We also describe an algorithm for the detection of bitangents on pairs of convex, disjoint image regions. While a previously published algorithm has a complexity of O(n) the expected time used by the proposed algorithm is of the order O(log(n)^2). These algorithms are used for the detection of pictographs with an alignment scheme. The number of generated hypotheses is reduced by selecting a set of stable model features in a training phase. Experiments show that the proposed strategy of feature selection allows for a very good recognition accuracy with a significantly reduced computational effort.