Appearance-based methods for mapping and localization have gained increasing attention in recent years. The strength of these models lies in their ability to represent the environment through high-level image features. However, the environment illumination, occlusions and walking people have a negative impact on these approaches. This paper presents a probabilistic appearance-based mapping and localization approach which uses the Feature Stability Histogram to update the environment appearance continuously and extract the more stable features in the environment. Our proposed method uses these stable features as successive appearance measurements to update the posterior probabilities incrementally on a topological map using a Rao-Blackwellized particle filter. Our algorithm considers omnidirectional images and laser data as measure of the environment appearance. Our approach was evaluated on a robot in a dataset collected along various seasons and time of day.