CBIR for urban building using A-KAZE features

Content based image retrieval systems are an active research topic in the Computer Vision domain. These systems are important when dealing with urban environment augmented reality applications like building landmarks recognition, tourism storytelling, cultural heritage and so on. In this paper we propose a content-based image retrieval system that handles recognizing buildings from an urban scenario using only visual cues. The system use a Bag of Features feature descriptor framework and, for extracting points of interest, image features. The proposed system is a natural step forward as Bag of Features or A-KAZE both have shown good results in other applications. To evaluate our proposed system, we have used the popular dataset, ZuBuD, where we obtained a 99% accuracy in detection.