The Influence of Local Descriptors on Content Based Image Retrieval Systems for Building Recognition

In this paper we will present our experiments regarding Content Based Image Retrieval for a public database of images containing buildings. We have based our research on the possible improvements brought by splitting the image into regions and computing a certain descriptor for each one of these. Our goal was to observe the influence of local descriptors on buildings recognition. Experimental tests performed on a public database of images containing buildings showed that the proposed approach provides significant improvements in the retrieval performance, compared to the global descriptors.