*2.4. Homomorphic Filter*

In order to solve the localization task, typical situations may happen such as lighting variations and changes in the position of some objects (chairs, tables, open doors, etc.). Hence, descriptors must be robust against these circumstances.

Fernandez et al. [41] showed that some pre-treatments could improve the localization accuracy in indoor environments with different lighting levels. Among the studied techniques, the use of the homomorphic filter [42] can be highlighted. The homomorphic filter permits filtering the luminance and reflectance components from an image separately.

The use of this filter has proven to provide especially good results when it is used in combination with the HOG descriptor [31], whereas in the FS and *gist* cases, the results were similar to or worse than without this pre-treatment filter. Hence, in the present paper, the following configurations will be used throughout the experiments: FS without filter, HOG with filter, and *gist* without filter.
