The strength of appearance-based mapping models lies in their ability to represent the environment through high-level image features; and provide humanreadable information. However, developing localization and mapping methods with these models could be very challenging, especially if robots must deal with long-term mapping, localization, navigation, occlusions, and dynamic environments. This paper proposes an appearance-based mapping and localization method based on the human memory model, which is used to build a Feature Stability Histogram (FSH) at each node in the robot topological map, these FSH register local feature stability over time through a voting scheme, and most stable features are considered for mapping and Bayesian localization. Experimental results are presented using omnidirectional images acquired through long-term acquisition considering: illumination changes (day time and seasons), occlusions, random removal of features, and perceptual aliasing. This method is able to adapt the internal node's representation through time to achieve global and local robot localization.