This work shows the design, development and evaluation of an automatic object recognition system, which implements a classification methodology that makes it possible to recognize objects from different contexts and environments. The evaluated categories are input as digital images by users through a mobile application. This system is also able to learn new categories and enhance its performance according with the user's feedback. The proposed architecture achieved an initial generalization error of 20% on the first 50 categories and we showed that this error decreased after a few days of interaction with users. Finally, the system learned three completely new categories with images collected through the mobile app.